Strike: v3.3 documentation Build date: 10-May-2002 -------------------------------------------------------------------------------- Documentation version: v1.3 Date: 26 September 2000 -------------------------------------------------------------------------------- Strike performs Groom-Bailey type tensor decomposition on magnetotelluric data from one or more sites in a specified band of frequency. The user may give a wide band for study, then spcify a smaller bandwidth within that wide band with or without an overlap. The MT data are to be in J-format, with one file per station. The code will read in a single site of data (required extension is ".dat"), or a list of data files. The user can specify an impedance relative error floor. If errors read in are less than this floor, then the errors are set to that floor. The floor is set on impedance, so is consistent between resistivity and phase. If the floor is set to zero, then the errors in the data files are used. NOTE: Comparison between parametric error estimates using Stodt's algorithms (the basis of Geotools errors and most contractors) and jackknife error estimates demonstrated that the parameteric errors are too small, by factors of up to ten, when the numbers of estimates is large (Chave and Jones, 1997). This typically occurs in the MT "dead-band", of 10 Hz - 10 s, and may explain the often poor fits one sees in the literature at these frequencies when decomposition procedures are applied. The user can specify whether realizations should be generated or not, and how many. This is for determining statistical bounds on the decomposition parameters. A bootstrap approach is used whereby new impedance realization estimates are generated using a random number generator and the variances of the impedance estimates. The variances on the decomposition estimates are derived either using a parametric approach (if the number of realizations is less than a specified number) or a jackknife approach. If the number of realizations is too low, then the extremal values are listed. The output files contain the decomposition parameters and their standard errors (or extremal values). This file can be converted to a J-format file using the supplied code "dcmp2j.f". -------------------------------------------------------------------------------- NAG routines The current version of strike uses NAG F77 routines. You have to provide these yourself. Information is available on web site www.nag.com. The main minimization routine in v3.1 of strike is E04UPF (html file included in distribution). This routine is available up to and including Version 18 of the F77 NAG library. In version 19 it was replaced by E04UNF (PDF file included in distribution). If you buy ver 19 I can provide you with ver 16 routines for comparison testing. (You will need to provide proof of ownership of a NAG licence before I can send you the ver 16 routines.) -------------------------------------------------------------------------------- QUESTIONS: >>>>Expert mode (y/n) [default: N ] > Unless your name is either Gary McNeice or Alan Jones, you should answer NO to this question. Answering YES will allow you to request estimates to be written out from the realizations, give to control over certain NAG minimization parameters, allow you to specify the starting solution and the starting seed integer. Unless you know NAG routines well, we advise you to leave well alone. The other parameters may be useful. >>>>Site list (.dat for single site)? [default: site.lis ] > Specify here either the name of a single data file (must end in ".dat"), or the name of a file which lists the data files you want to use. The data are to be in J-format. >>>>Give impedance relative error floor (in %) [default: 1.7500 ] > Specify impedance relative error floor. A value of 1.75% in impedance is equivalent to 1 deg in phase and 3.52% in apparent resistivity. >>>>>Give normalization type (? for list) [default: GAVSD2 ] > ***NEW to v3.1*** Sets the denominator in the minimization scheme. L2 L2 unweighted normalization MAXSD maximum s.d. weighted normalization GAVSD geom. av. s.d. weighted normalization GAVSD2 geom. av. s.d. (Zxx,Zyy) & (Zxy,Zyx) weighted normalization SUMSQ sqrt(sum squares s.d.) weighted normalization SUMSQ2 sqrt(sum squares s.d. (Zxx,Zyy) & (Zxy,Zyx)) weighted normalization It is useful to try different normalizations to see how robust the strike angle determined is. If it varies greatly with different schemes, then the data probably have different strikes over the frequency/site ranges chosen. The default seems to give the most consistent results. >>>>Enter minimum period? [default: 0.10000E-03 ] > Specify the minimum period to study >>>>Enter maximum period? [default: 10000. ] > Specify the maximum period to study >>>>Enter bandwidth (no. of period decades)? [default: 2.0000 ] > Specify the bandwidth of the bands that you wish to study. The default is the whole bandwidth between the minimum and maximum periods specified. If you give a value any less than this, then you will be working with multiple bands, each of which will derive independent decomposition parameters. This is a good way of studying depth-dependent strike variation. >>>>Enter overlap (no. of period decades)? [default: 0.0 ] > Specify the overlap between bands. The default is for no overlap. >>>>Place bounds on parameters (T or F) ? [default: Y ] > Specify whether you wish to bound the decomposition parameters or not. >>>Change bounds from standard bounds (y/n)? [default: N ] > The standard bounds are Regional azimuth bounds -360.000 360.000 Shear bounds -45.0000 45.0000 Twist bounds -60.0000 60.0000 You can specify whether you wish to change these bounds if you wish a constrained fit. NOTE: Often the minimization routine will hit a bound. Once there because the default start for minimization is WARM (i.e., use the last parameters found as the new starting solution) then it will stay at the bound. >>>>Do statistics (y/n) [default: N ] > Specify whether you wish multiple-realizations to be generated so that the bounds on the parameters will be determined using either a parametric approach (for number of realizations less than a given minimum number) or a jackknife approach to the bootstrap-generated estimates. If you say YES, then you will be asked: >>>>Give number of realizations (default is maximum permitted) [default: 21 ] > Specify the number of realizations to generate. -------------------------------------------------------------------------------- size.inc The user can modify the dimension sizing to be appropriate for the problem by editing size.inc. The parameters should be set as small as possible, as the requrements of NAG workspace go as the square of the number of data. There are six parameters that are user-definable, and are: MAXS maximum number of sites MAXF maximum number of frequencies per band MAXDAT maximum number of data frequencies in file MAXBND maximum number of bands NREL maximum number of realizations MINREL minimum number of realizations for jackknife statistics None of the other parameters should be modified. The executable becomes really large for large problems. I think I have the parameters set properly for the NAG routine, but I may have something set incorrectly. If you note an error in the size.inc file, please let me know. -------------------------------------------------------------------------------- OUTPUT: The output files are one for each station with the name the same as the input file and the extension ".dcmp" (".dcm" on DOS). This file contains the decomposition parameters and their ranges. The parameters are listed in the order: 14 regional azimuth 14 shear angle 14 channelling angle 14 twist angle 14 app rho a 14 app rho b 14 imped phase a 14 imped phase b 14 error 14 skew 14 anis where the integer ("14" in this case) is the number of periods. Various codes exist for plotting and reformatting these .dcmp files. Also, the output files can be converted to J-format using "dcmp2j". Once in J-format, they can be converted to EDI using "j2edi" available from MTNet. -------------------------------------------------------------------------------- EXAMPLE: An example data file is supplied, strike_example.dat. It contains 25 estimates in the range 4 - 13653 s. The run is also supplied, (strike_example.run) as is the main output file (strike_example.dcmp). -------------------------------------------------------------------------------- CONDITIONS OF USE: These are the conditions that you agreed to when you obtained the code: 1. I accept the McNeice-Jones multi-site, multi-frequency magnetotelluric tensor decomposition code, hereinafter called "strike", on a caveat emptor basis. 2. I will use strike for non-profit purposes only. 3. I will not accept any payment for use of strike. 4. I will not give strike to any other person. 5. I will inform the authors of strike, Gary McNeice and Alan Jones, of any coding errors that I find. 6. I will inform the authors of strike, Gary McNeice and Alan Jones, of any improvements and additions that I make. 7. I will acknowledge the use of strike in any publications and presentations I write or give in which I present results based on strike-based decomposition or decomposed data. 8. I understand that the current version of strike, v3.0, uses NAG routines that I have to provide myself. -------------------------------------------------------------------------------- REFERENCE TO USE: The current reference to use is: McNeice, G. and A.G. Jones, 2001. Multisite, multifrequency tensor decomposition of magnetotelluric data. Geophysics, 66, in press. -------------------------------------------------------------------------------- REFERENCES: Chave, A.D. and A.G. Jones, 1997. Electric and magnetic field distortion decomposition of BC87 data. J. Geomagn. Geoelectr., 49, 767-789.