Mathematical operations on grids in the spectral domain


gmt grdfft ingrid [ ingrid2 ] -Goutfile|table [ -Aazimuth ] [ -Czlevel ] [ -D[scale|g] ] [ -E[r|x|y][+n][+w[k]] ] [ -F[r|x|y]params ] [ -I[scale|g] ] [ -Nparams ] [ -Q] [ -Sscale|d ] [ -V[level] ] [ -fflags ] [ --PAR=value ]

Note: No space is allowed between the option flag and the associated arguments.


grdfft will take the 2-D forward Fast Fourier Transform and perform one or more mathematical operations in the frequency domain before transforming back to the space domain. An option is provided to scale the data before writing the new values to an output file. The horizontal dimensions of the grid are assumed to be in meters. Geographical grids may be used by specifying the -fflags option that scales degrees to meters. If you have grids with dimensions in km, you could change this to meters using grdedit or scale the output with grdmath.

Required Arguments


Optionally, append =ID for reading a specific file format [Default is =nf] or ?varname for a specific netCDF variable [Default is the first 2-D grid found by GMT]. The following modifiers are supported:

  • +b - Select a band [Default is 0].

  • +d - Divide data values by the given divisor [Default is 1].

  • +n - Replace data values matching invalid with NaN.

  • +o - Offset data values by the given offset [Default is 0].

  • +s - Scale data values by the given scale [Default is 1].

Note: Any offset is added after any scaling.


Specify the name of the output grid file (see Grid File Formats) or the 1-D spectrum table (see -E).

Optional Arguments


Take the directional derivative in the azimuth direction measured in degrees CW from north.


Upward (for zlevel > 0) or downward (for zlevel < 0) continue the field zlevel meters.


Differentiate the field, i.e., take \(\frac{\partial}{\partial z}\) of the grid z. This is equivalent to multiplying by \(k_r\) in the frequency domain (\(k_r\) is radial wave number). Append a scale to multiply by \(k_r \cdot\)scale) instead. Alternatively, append g to indicate that your data are geoid heights in meters and output should be gravity anomalies in mGal. Repeatable. [Default is no scale].


Estimate power spectrum in the radial or a horizontal direction. No grid file is created. If one grid is given then f (i.e., frequency or wave number), power[f], and 1 standard deviation in power[f] are written to the file set by -G [standard output]. If two grids are given we write f and 8 quantities: Xpower[f], Ypower[f], coherent power[f], noise power[f], phase[f], admittance[f], gain[f], coherency[f]. Each quantity is followed by its own 1-std dev error estimate, hence the output is 17 columns wide. Select your spectrum by choosing one of these directives:

  • r - Choose a radial spectrum [Default].

  • x - Compute the spectrum in the x-direction instead.

  • y - Compute the spectrum in the y-direction instead.

Two modifiers are available the adjust the output further:

  • +w - Write wavelength w instead of frequency f, and if your grid is geographic you may further append k to scale wavelengths from meter [Default] to km.

  • +n - Normalize spectrum so that the mean spectral values per frequency are reported [By default the spectrum is obtained by summing over several frequencies.


Filter the data. Place x or y immediately after -F to filter x or y direction only; default is isotropic [r]. Choose between a cosine-tapered band-pass, a Gaussian band-pass filter, or a Butterworth band-pass filter.


Specify four wavelengths lc/lp/hp/hc in correct units (see -fflags) to design a bandpass filter: wavelengths greater than lc or less than hc will be cut, wavelengths greater than lp and less than hp will be passed, and wavelengths in between will be cosine-tapered. E.g., -F1000000/250000/50000/10000 -fflags will bandpass, cutting wavelengths > 1000 km and < 10 km, passing wavelengths between 250 km and 50 km. To make a highpass or lowpass filter, give hyphens (-) for hp/hc or lc/lp. E.g., -Fx-/-/50/10 will lowpass x, passing wavelengths > 50 and rejecting wavelengths < 10. -Fy1000/250/-/- will highpass y, passing wavelengths < 250 and rejecting wavelengths > 1000.

Gaussian band-pass:

Append lo/hi, the two wavelengths in correct units (see -fflags) to design a bandpass filter. At the given wavelengths the Gaussian filter weights will be 0.5. To make a highpass or lowpass filter, give a hyphen (-) for the hi or lo wavelength, respectively. E.g., -F-/30 will lowpass the data using a Gaussian filter with half-weight at 30, while -F400/- will highpass the data.

Butterworth band-pass:

Append lo/hi/order, the two wavelengths in correct units (see -fflags) and the filter order (an integer) to design a bandpass filter. At the given cut-off wavelengths the Butterworth filter weights will be 0.707 (i.e., the power spectrum will therefore be reduced by 0.5). To make a highpass or lowpass filter, give a hyphen (-) for the hi or lo wavelength, respectively. E.g., -F-/30/2 will lowpass the data using a 2nd-order Butterworth filter, with half-weight at 30, while -F400/-/2 will highpass the data.

Note: For filtering in the time (or space) domain instead, see grdfilter.


Filename for output netCDF grid file OR 1-D data table (see -E). This is optional for -E (spectrum written to standard output) but mandatory for all other options that require a grid output.


Integrate the field, i.e., compute \(\int z(x,y) dz\). This is equivalent to divide by \(k_r\) in the frequency domain (\(k_r\) is radial wave number). Append a scale to divide by \(k_r \cdot\)scale instead. Alternatively, append g to indicate that your data set is gravity anomalies in mGal and output should be geoid heights in meters. Repeatable. [Default is no scale].


Choose or inquire about suitable grid dimensions for FFT and set optional parameters. Control the FFT dimension via these directives:

  • a - Let the FFT select dimensions yielding the most accurate result.

  • f - Force the FFT to use the actual dimensions of the data.

  • m - Let the FFT select dimensions using the least work memory.

  • r - Let the FFT select dimensions yielding the most rapid calculation.

  • s - Just present a list of optional dimensions, then exit.

Without a directive we expect -Nnx/ny which will do FFT on array size nx/ny (must be >= grid file size). Default chooses dimensions >= data which optimize speed and accuracy of FFT. If FFT dimensions > grid file dimensions, data are extended and tapered to zero.

Control detrending of data by appending a modifier for removing a linear trend. Consult module documentation for the default action:

  • +d - Detrend data, i.e. remove best-fitting linear trend.

  • +a - Only remove the mean value.

  • +h - Only remove the mid value, i.e. 0.5 * (max + min).

  • +l - Leave data alone.

Control extension and tapering of data by appending a modifier to control how the extension and tapering are to be performed:

  • +e - Extend the grid by imposing edge-point symmetry [Default].

  • +m - Extends the grid by imposing edge mirror symmetry.

  • +n - Turns off data extension.

Tapering is performed from the data edge to the FFT grid edge [100%]. Change this percentage via modifier +twidth. When +n is in effect, the tapering is applied instead to the data margins as no extension is available [0%].

Control messages being reported:

  • +v - Report suitable dimensions during processing.

Control writing of temporary results: For detailed investigation you can write the intermediate grid being passed to the forward FFT; this is likely to have been detrended, extended by point-symmetry along all edges, and tapered. Use these modifiers to ave such grids:

  • +w - Set the suffix from which output file name(s) will be created (i.e., ingrid_prefix.ext) [Default is “tapered”], where ext is your file extension

  • +z - Save the complex grid produced by the forward FFT. By default we write the real and imaginary components to ingrid_real.ext and ingrid_imag.ext. Append p to instead use the polar form of magnitude and phase to files ingrid_mag.ext and ingrid_phase.ext.


Selects no wavenumber operations. Useful in conjunction with -N modifiers when you wish to write out the 2-D spectrum (or other intermediate grid products) only.


Multiply each element by scale in the space domain (after the frequency domain operations). [Default is 1.0]. Alternatively, append d to convert deflection of vertical to micro-radians.


Select verbosity level [w]. (See full description) (See cookbook information).


Geographic grids (dimensions of longitude, latitude) will be converted to meters via a “Flat Earth” approximation using the current ellipsoid parameters.

-^ or just -

Print a short message about the syntax of the command, then exit (Note: on Windows just use -).

-+ or just +

Print an extensive usage (help) message, including the explanation of any module-specific option (but not the GMT common options), then exit.

-? or no arguments

Print a complete usage (help) message, including the explanation of all options, then exit.


Temporarily override a GMT default setting; repeatable. See gmt.conf for parameters.

Grid Distance Units

If the grid does not have meter as the horizontal unit, append +uunit to the input file name to convert from the specified unit to meter. If your grid is geographic, convert distances to meters by supplying -fflags instead.


netCDF COARDS grids will automatically be recognized as geographic. For other grids geographical grids were you want to convert degrees into meters, select -fflags. If the data are close to either pole, you should consider projecting the grid file onto a rectangular coordinate system using grdproject

Data Detrending

The default detrending mode is to remove a best-fitting linear plane (+d). Consult and use -N to select other modes.

Normalization of Spectrum

By default, the power spectrum returned by -E simply sums the contributions from frequencies that are part of the output frequency. For x- or y-spectra this means summing the power across the other frequency dimension, while for the radial spectrum it means summing up power within each annulus of width delta_q, the radial frequency (q) spacing. A consequence of this summing is that the radial spectrum of a white noise process will give a linear radial power spectrum that is proportional to q. Appending n will instead compute the mean power per output frequency and in this case the white noise process will have a white radial spectrum as well.


Note: Below are some examples of valid syntax for this module. The examples that use remote files (file names starting with @) can be cut and pasted into your terminal for testing. Other commands requiring input files are just dummy examples of the types of uses that are common but cannot be run verbatim as written.

To obtain the normalized radial spectrum from the remote data grid, after removing the mean, let us try:

gmt grdfft -Er+n -N+a > spectrum.txt

To upward continue the sea-level magnetic anomalies in the file to a level 800 m above sealevel:

gmt grdfft -C800 -V

To transform geoid heights in m ( on a geographical grid to free-air gravity anomalies in mGal:

gmt grdfft -Dg -V

To transform gravity anomalies in mGal ( to deflections of the vertical (in micro-radians) in the 038 direction, we must first integrate gravity to get geoid, then take the directional derivative, and finally scale radians to micro-radians:

gmt grdfft -Ig -A38 -S1e6 -V

Second vertical derivatives of gravity anomalies are related to the curvature of the field. We can compute these as mGal/m2 by:

gmt grdfft -D -D -V

To compute cross-spectral estimates for co-registered bathymetry and gravity grids, and report result as functions of wavelengths in km, try:

gmt grdfft gravity.grd -E+wk -fg -V > cross_spectra.txt

To examine the pre-FFT grid after detrending, point-symmetry reflection, and tapering has been applied, as well as saving the real and imaginary components of the raw spectrum of the data in, try:

gmt grdfft -N+w+z -fg -V -Q

You can now make plots of the data in,, and

See Also

gmt, grdedit, grdfilter, grdmath, grdproject, gravfft