[Shake-dev] intensity bias in ShakeMap v3.5

[Bruce Worden]

Hi Georgia,

There was a bug in Wald99 (and AK07) where the intensities returned  
were allowed to exceed the 1 to 10 range. Oddly, the limiting code was  
in there, but commented out -- no doubt an error on my part. Fixing  
these bugs (and a minor unrelated bug in grind) now produces the  
expected result: if all of your intensities are 1, your bias will be  
zero.

It is important to use converted PGM in the intensity bias because not  
all maps have observed intensity, and without some data we can't do  
the bias. Doing the bias results in a significant reduction in the  
error of the map. I've tested this extensively both with and without  
observed intensities. In cases where you do have observed intensities,  
those are weighted higher than the converted PGMs in the bias  
calculation.

You can get the updated code with 'svn update' and 'make'. Let me know  
how it goes.

Bruce


On Jan 18, 2011, at 3:19 AM, Georgia Cua wrote:

> Dear Bruce, shake-dev,
>
> I have been working on our ShakeMap v3.5 installation in Switzerland.
>
> I noticed something peculiar with the behaviour of the intensity bias.
>
> I have the observed PGA and PGV from a M3.3 in Switzerland. (To be  
> safe, I used Small as the GMPE and Wald99 as the GMICE. The  
> behaviour is similar when I use our Swiss-specific GMPE and GMICE.)
>
> When I run this event with the observed PGM as input, there are  
> large bias values for PGA and PGV, -1.5 and -0.7. (Those are  
> expected, as PGA and PGV in Switzerland tend to be lower for a given  
> magnitude than predicted by California relationships.) What troubles  
> me is that the intensity magnitude bias turns out positive (1.43).
>
> If I run a shakemap for this event without the PGM as input (just  
> providing the event.xml file), the maximum intensity is 3.64. When I  
> run it with the PGM as input, the maximum intensity is 5.1, even  
> though the PGA and PGV biases are both negative.
>
> I've run this same event on our v3.2 installation, and there, it  
> seems to behave as I expect, where if the PGA and PGV biases are  
> negative, the maximum intensity using the PGM observations will be  
> lower than when running it without the observations.
>
> I am attaching the message logs from running shake, grind.conf, the  
> info.xml file, as well as the regression plots.
>
> I am not sure where I am going wrong. Thanks in advance for your help.
>
> Georgia
>
> PS. When there is no native intensity data provided as input, and no  
> IPE is specified, should the intensity bias still calculated?
>
> < 
> test33 
> .with_pgm 
> .shake 
> .log 
> > 
> < 
> ATT00001 
> .html 
> > 
> < 
> grind 
> .conf 
> > 
> < 
> ATT00002 
> .html 
> > 
> < 
> with_pgm 
> .info 
> .xml 
> > 
> < 
> ATT00003 
> .html 
> > 
> < 
> with_pgm 
> .pga_regr 
> .png 
> > 
> < 
> ATT00004 
> .html 
> > 
> < 
> with_pgm 
> .pgv_regr.png><ATT00005.html><with_pgm.mi_regr.png><ATT00006.html>
> _______________________________________________
> Shake-dev mailing list
> Shake-dev at geohazards.usgs.gov
> https://geohazards.usgs.gov/mailman/listinfo/shake-dev