[Shake-dev] ShakeMap update, changes to site amplification

[Bruce Worden]

Hi Folks,

(Note: For those of you who received an earlier version of this email, there have been three more changes (see the last three bullet points), so you'll need to do svn update again to get them.)

Sorry about the very long email, but there are some significant changes...

I've just updated the ShakeMap code repository. The most significant changes are to the way site amplifications are computed when you are NOT using the GMPE-native site corrections (i.e., you are not running grind with -gmpesc or -nativesc). The new approach is, I believe, more consistent with Borcherdt's 1994 paper, and is actually simpler and more flexible. I will describe it below. First, though, I have more or less preserved the old approach (with a couple of minor exceptions), and it can be accessed by running grind with the -oldsc flag. The differences are:

1) In previous versions, PGA was used as the input ground motion to select which amplification factors to apply to ALL of the ground-motion parameters -- this was partly for convenience and partly a bug (at least where the PSA amps were concerned).

2) The structure of the Borcherdt functions applies to accelerations, not velocity (this is part of the reason for #1, above). The updated code now converts PGV to PSA10 via Newmark and Hall (1982), then amplifies (or de-amplifies) that, then converts it back to PGV. The Newmark/Hall conversion is completely linear, so there is no loss of information, and we've always treated PGV as interchangeable with PSA10 (see the HazusPGV GMPE module, for example), so I think this is reasonable to do.

That's all I'm going to say about the old approach (accessed via -oldsc).

The NEW approach uses Borcherdt's equations 7a and 7b directly:

Fa = (v0/v)^Ma
Fv = (v0/v)^Mv

for the short- and mid-period amplification factors. Ma and Mv vary by input ground motion, but my understand is that the values given in the paper are just picked off a linear fit of empirical and modeling results. So they are actually meant to be continuous, not stepwise, as they were implemented in the old version. 

Implementing them as smoothly varying parameters prevents the "ringing" effect that happens when a small drop in ground motion causes a jump to a new, higher, amplification factor and, thus, a jump in amplitude above the previous level. In addition, the new approach allows a continuous variation in Vs30, rather than the fixed steps of the old approach. 

The new approach requires a new (simpler) amplification table. I've created a default one using Borcherdt's Ma and Mv factors from Table 2 in his paper, but operators are invited to develop their own for their own regions. The system tries to distinguish the old tables from the new, so if you configure ampfactor_file with an old table and don't use -oldsc (or vice-versa), grind will spit out a warning and try to do something sensible.

This is all discussed in a bit more detail in a new section (3.4.4) of the Software Guide, which I recommend everyone read.

Another major change is in the way MMI site corrections are handled. Previously, the operator had to run makeMMIsiteTable to create tables for MMI site amplification, and configure mifactor_file in grind.conf. This has all disappeared, and is now handled internally by grind: Intensities are converted to PGA and PGV using the GMICE, amplified using either the new or old ground-motion site amplifications (described above), and then converted back to intensity. 


SUMMARY:

If you use the GMPE-native site corrections (via -gmpesc to grind):

o Nothing has changed, but you should remove mifactor_file from grind.conf to prevent a run-time error (assuming you have configured it).

If you don't use -gmpesc:

o If you do nothing, you will be quietly switched to the new site amplification functions using the new table <shake_home>/lib/sitecorr/Borcherdt94.dat (though you may also create and specify your own table).

o If you run grind with -oldsc, you will continue to use the old functions and site correction tables, with only minor differences (improvements).

o No matter which of the non-GMPE-native site amplification systems you choose, the MMI site amplifications are now done the new way, and you no longer need to create amp tables with makeMMIsiteTable. You may delete the MMI tables you created, and remove the parameter "mifactor_file" from grind.conf.



Other changes in this release:

o grind.conf has been updated with new doc for ampfactor_file, and mifactor_file has been removed. I've taken this opportunity to get rid of old comments about obsolete parameters. All users should remove legacy 'mifactor_file' lines in grind.conf.

o Updated the IPE TA09_mmi with the new coefficients in the latest version of the paper. Site corrections, however, are still handled the way they were in the previous version.

o Added ESRI .lyr files to the GIS zip files.

o Fix to JpgMaps.pm to better crop the intensity overlay (per Licia Faenza). Operators, please verify that your KML overlays are correct.

o Updated software guide; added a PDF version of the Guide.

o Improved reporting of site corrections in info.xml.

o Changed adhoc.lis to adhoc.lis.CISN to avoid overwriting other networks' adhoc.lis with each update.

o db2xml will no longer flag any stations based on database flags on Z channels (via Pete Lombard).

o Added a new output file "uncertainty.xml." Similar in structure to grid.xml, the new file provides the calculated one-sigma uncertainty at each grid point for each parameter. The file will appear on the Downloads page of the web site. Updated the About Formats page with info about the new file, as well as revised descriptions of grid.xml.

o Fix to Source.pm to set the zone attribute correctly (via Pete Lombard).

o Changes to topo2grd: Upped the VS30_OCEAN to 600 (m/s), added VS30_MINIMUM and set it to 180 (m/s).

o IMPORTANT: WGRW11 is now the default GMICE -- if you've been relying on the default settings to get Wald99 and want to keep using it, you need to explicitly configure it with the 'mi2pgm' and 'pgm2mi' parameters in grind.conf. WGRW11 is a new GMICE based on an upcoming BSSA paper by Worden, Gerstenberger, Rhoades, and Wald (manuscript available upon request). It is very similar to Wald99 at the upper end, but is more consistent with the data for lower ground motions. It also has distance and magnitude correction terms, and has relationships for the PSA parameters. Like Wald99, it was developed with California data only, so you will want to check its applicability to your region before using it in production.

o The spatial correlation function (see Worden, et al (2010)) now defaults to Goda and Atkinson (2010), rather than being based on Boore, et al. (2003). The G&A work was much more comprehensive than Boore, and is frequency dependent, so we feel the change is justified. However, I have made the choice configurable. If you want the Boore function, you can edit <shake_home>/src/contour/Makefile and change CORR_FUNC from ga10 to boore03, then run 'make'. You can also write your own spatial correlation function -- see boore03.c and ga10.c for examples. If you do change the Makefile, keep an eye on it when you upgrade in the future, as it may switch back to ga10.

o The intermediate segments separating quadrilaterals in multi-quadrilateral polygonal faults are now plotted (as dashed lines).

As always, let me know if you run into problems. I realize that I may not have explained the site amplification stuff very well, so feel free to ask.

Bruce

P.S. It is fairly easy to compare new vs. old site amplifications: Leave your ampfactor_file configured to your old Borcherdt table and run with -oldsc. Make your maps and set them aside. Then run grind without -oldsc -- grind will complain that your ampfactor_file is incorrect, but it will then default to the new Borcherdt94.dat file. My experience is that the differences are not large.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://geohazards.usgs.gov/pipermail/shake-dev/attachments/20111011/916f04a9/attachment.html>