Data and model files

MTGeophysics.jl reads and writes the same files as ModEM. This means you can prepare data with your usual tools, run a model here, and continue in ModEM, or the other way around. In this section we look at the files you will meet most often.

Data files

1D, 2D and 3D all use one data format: the ModEM Full_Impedance file. Impedances are in [mV/km]/[nT] with the exp(+iωt) sign convention. 3D files may also hold the tipper.

using MTGeophysics

d3 = load_data_modem("data.dat")   # the full file: impedance tensor and tipper
d2 = load_data2d("data.dat")       # the ZXY and ZYX parts, as 1D and 2D use them

In 2D, ZXY is the TE mode and ZYX the TM mode. This only holds when the data are in the strike frame, so every 2D workflow first rotates the data to the strike (see Rotation and strike).

Note

A data file whose impedances are all zero works as a template. A forward run fills it with the predicted response for those stations and frequencies.

EDI files

Field data usually come as SEG EDI files, one per site. They are converted to a ModEM data file once, and everything else works on that file:

EDIToModEM("EDI/COPROD2")                  # a directory of EDIs -> EDI/COPROD2.dat
EDIToModEM("site.edi", "site.dat")         # one file
ModEMToEDI("data.dat")                     # one EDI per site -> data_EDI/

EDIToModEM reads the three EDI layouts in use: transfer functions (>ZXXR … >TYVAR.EXP, errors the square root of the variances), apparent resistivity and phase only (>RHOXY, >PHSXY), and cross spectra (>=SPECTRASECT). From spectra, Z and T are formed with the remote-reference channels. Their errors are the 68 % bounds of the local regression, from the residual power and the magnetic autopowers. They are an estimate and differ from the errors the acquisition processing reports.

Periods are shared between sites within 0.2 %. Station x and y are metres in a transverse Mercator projection about the sites' centroid, as in MakeMesh3D. Site names come from DATAID, or from SECTID when all files share one DATAID. z_floor and t_floor raise the errors on the way. A file that cannot be read is skipped with a warning.

Note

Some published EDIs are a unit factor off. EDIToModEM warns and rescales a file whose apparent resistivities come out beyond 10⁶ Ω·m, or below 10⁻³ Ω·m, when one [mV/km]/[nT] ↔ Ohm factor brings them back. The MTNet BC87 files need this. fix_units = false turns it off.

A ModEM file holds one rotation. If the EDIs are in different frames (ZROT), every site is turned to north on the way, and a site-period without the full tensor is dropped.

Model files

Dimension	Format	Read	Write
1D	the 2D layout with a single column	ReadModel2D	WriteModel2D
2D	ModEM 2D layout, ln ρ	ReadModel2D	WriteModel2D
3D	WS3D / ModEM 3D	load_ws3d_model, load_model_modem	write_ws3d_model, write_model_modem

Cells above the ground are air. They are written as 10¹⁷ Ω·m, and any cell above 10¹⁵ Ω·m is read back as air. This is how topography is stored in a model file.

Control files

The 1D and 2D workflows are driven by short text files with one key : value per line, for example:

Mode                     : TETM
Strike (deg)             : auto

File	Used by	What it sets
fwd.ctrl	2D forward and all 2D inversions	mode, strike and the air layers
inv.ctrl	2D Gauss–Newton and NLCG	algorithm, regularisation, stopping rules
vfsa.ctrl	2D VFSA	search bounds, chains, cooling
cov.ctrl	2D Gauss–Newton and NLCG	which cells are free, fixed, air or water
mask.ctrl	2D VFSA	the same mask as cov.ctrl, without a covariance
InvCtrl.GN, InvCtrl.VFSA	1D inversion	algorithm and stopping rules
topo.dat	2D mesh tool	ground elevation along the profile

Ready-to-use examples are in examples/ctrl/1D and examples/ctrl/2D, and the 2D mesh tool writes a full set for new data. Every key is listed in the control file reference.

Misfit

The misfit between observed and predicted data is reported as RMS: the square root of χ² divided by the number of real data. The real and imaginary parts of each impedance count as two data.

fit = chi2_rms2d(load_data2d("data.dat"), load_data2d("data.pred"))   # 1D and 2D
fit = chi2_and_rms("data.dat", "pred.dat")                            # 3D
fit.rms