Step 22: Score the model against data

The testable predictions of any Shells model include:

 relative velocities of geodetic benchmarks;

 most-compressive horizontal principal stress azimuths;

 long-term fault heave and throw rates;

 rates of seafloor spreading;

 the distribution of seismicity on the map;

 fast-polarization directions of split SKS arrivals.

To compare any combination of these data sets with your model, run my program OrbScore2.
If you use Windows (95, 98, NT, 2000, ME, XP, Vista), then run OrbScore2_batch.exe (supplied) for Windows.
If you don’t use Windows, then compile your own version from OrbScore2.f90 to use with other operating systems.

If you compiled OrbScore2.f90, with switches set to give interactive prompting for undefined filenames at run time, you can specify input and output file names as OrbScore2 runs. Otherwise, you will have to predefine the connections between input/output filenames and Fortran I/O device numbers, using a batch file.
Here is a sample of the format (and possible contents) of a Windows batch file: OrbScore2.bat

Here is the complete list of input and output channels for OrbScore2. (If a particular device number causes problems on your system, you can change the corresponding IUNITx parameter and recompile.) The files marked in yellow are always required. If some of these scoring data sets (channels 11 to 15 and 32 to 34) are not available, just supply non-existent file names, or the names of empty files. OrbScore2 will skip to the next category of scoring data.

Fortran device number (channel)

Type of input/output file to connect:

Sample file of this type on my web site:

1
(or: IUNITG)

finite element grid (.feg) file [required input]

Earth5R.feg

2
(or: IUNITP)

parameter input (.in) file [required input]

iEarth5-049.in

3
(or: IUNITV)

nodal velocity (v_____.out) file [required input to OrbScore, although "output" from Shells]

vEarth5-049.out

6
(or: IUNITT)

scoring results (s_____.out) file [PRIMARY output]

sEarth5-049.out

11
(or: IUNITB)

geodetic benchmark positions and relative horizontal velocities, with uncertainties (in any velocity reference frame) [optional input]

GPS2006_selected_subset.gps; for format conventions, see gps_format.txt

12
(or: IUNITS)

most-compressive horizontal principal stress azimuths, with quality grades [optional input]

robust_interpolated_stress_for_OrbScore2.dat; for format conventions, see stress_format.txt.

13
(or: IUNITD)

long-term fault heave and/or throw rates from geologic data [optional input]

(none); for format conventions, see slip_rate_format.txt

14
(or: IUNITM)

sea-floor spreading rates (full rate, not half-rate) at mid-ocean ridges [optional input]

magnetic_PB2002.dat

15
(or: IUNITE)

seismic catalog (.cat) file, for computation of seismic strain-rate map [optional input]

(none)

21
(or: IUNITO)

dummy .feg file, for use with FiniteMap to plot the smoothed seismic strain-rate map (produced only if input 15 is present) [optional output]

(none)

22
(or: IUNITZ)

dummy .feg file, for use with FiniteMap to plot the smoothed model strain-rate map (produced only if input 15 is present) [optional output]

(none)

23
(or: IUNITX)

special-case output, still under development [no output is likely]

(none)

31
(or: IUNITY)

errors in geodetic velocity predictions, in same .gps format as the input dataset.  Note that this occurs only IF (pltGEO) .AND. (some geodetic data are provided for scoring).  [optional output file]

(none)

32
(or: IUNITK)

upper-mantle anisotropy data, in the form of fast-polarization ("phi") azimuths from SKS splitting, together with SKS splitting times in s.  [optional input file]

Fouch_2004_SKS_splitting-selected.dat

33
(or: IUNITQ)

torque report (obtained by running Shells version 2006.08.26 or later).  This is used in scoring the anisotropy data read through IUNITK (above).  [optional input file]

qEarth5-049.out

34
(or: IUNITF)

outlines of the plates (e.g., PB2002_plates.dig).  This is used in scoring the anisotropy data read through IUNITK (above).  [optional input file]

PB2002_plates.dig

One caution: Up to this point, my programs have been free of units, so that any internally-consistent set of units can be used. (I prefer SI = mks.) However, OrbScore2 assumes that the .feg file, .in file, and v_____.out file are in SI (as before), while the geodetic, fault slip-rate, and sea-floor spreading data sets are in units of  mm/a (= mm/year). This is because it is just so inconvenient to look at tables of such numbers in m/s! If you need to change this convention, see parameter UNITL and UNITT in the code of OrbScore2.f90, and adjust these.

Now, look at the text output from Fortran unit 6. Tables are printed to compare each test datum with the corresponding model prediction, and at the end of each table a selection of summary error measures is provided (mean absolute value of error, RMS error, et cetera). At the very end of the output on Fortran unit 6, one of these scalar measures is reprinted, along with the three title lines for this model. (These titles were derived from the first lines of the .feg file, .bcs file, and .in file, respectively, and stored in the v_____.out file.) Once you are sure that OrbScore2 is working properly, you can just jump to this summary table to see how successful each model was.