[Met_help] MET questions

John Halley Gotway johnhg at rap.ucar.edu
Fri Sep 5 15:50:55 MDT 2008


Daniel,

Regarding your second question, you've got it exactly correct.  That was a good synopsis of the method.  Let me add one additional thing though.  The "nbr_threshold" parameter may be set to indicate
how many valid data points must be present to compute a "coverage" value.  If nbr_threshold is set to 1.0, then all n*n points must contain valid data - in your example all 9 points would have to be
valid.

If you set nbr_threshold = 0.75, only 7 out of the 9 points would need to contain valid data.  Suppose that's the case, 7 of the 9 contain valid data, and 4 of those 7 contain values which exceed the
threshold.  In this case, the coverage value would be computed as 4/7 = .5714, and not 4/9.  Make sense?

Regarding your first question, it's a good one.  Currently, grid_stat is set up to only compare fields of the same type.  And the same thresholds are applied to the fcst/obs fields.  I can't think of
an "easy" way for you to generalize this without getting your hands pretty messy in the code.  If it's possible to "rescale" one of the fields to match up well with the other, I'd suggest doing that.

We did consider this issue when writing the code, however at that time, we were limited by trying to conform to NCEP's VSDB format.  If we were going to apply different thresholds to the forecast and
observation fields, we would need to indicate that in the output somewhere.  However, adding more columns to the output format to hold that information would break the format - which is something we
may eventually do anyway.

The MODE tool IS set up to handle comparisons between different fields.  In the MODE configuration file, you can specify the fcst/obs fields separately and their thresholds separately.  And the MODE
output files contain additional columns to store that information.  And one output of MODE is a file containing contingency table counts.  But you wouldn't be able to run neighborhood methods though
MODE.  Those are only implemented in grid-stat.

Hope that helps,
John

Daniel Schaffer wrote:
> Hello (presumbably John)
> 
> We are thinking about running MET grid_stat to generate dichomomous counts from forecast and observed convection data in the next few weeks.  You may recall I had some email exchanges with you last January relating to this during some initial investigation I did.  Now, upon revisiting this, a couple of new questions have come up.
> 
> 1. From my reading of the MET 1.1 User's guide, it appears that I can specify a threshold that applies to both the forecast and obs fields.  However, I don't see how I can specify different thresholds for the forecast and obs.  For our convection data, the forecast and obs are not on the same scale.  Is there any easy code change I could make that would extend MET to handle this case?  My other option is to put the forecast and obs data on the same scale prior to running MET.
> 
> 2. I want to ensure I am clear on how the neighborhood analysis works.  The documentation on page 5-7 says that "forecast and observation coverage" fields are computed from the raw fields and a raw threshold value.  Suppose my nbr_width is 3 so I have 9 grid points including the center grid point of interest.  Then if 6 boxes exceed the threshold and 3 do not, the coverage value at that grid point is 2/3?  And then the coverage field is computed by repeating this exercise over all grid points?  And then, finally, the nbr_frac_threshold is use to create neighborhood contingency tables from the paired forecast/obs coverage grids in exactly the same way as in the case without neighborhoods.  So, for example, if at the gridpoint mentioned above, the forecast coverage is 2/3, the obs coverage .5 and the nbr_frac_threshold is 0.4 then the "Yes-Yes" count would be incremented?
> 
> Thanks,
> Dan Schaffer
> 
> -------------------------------------------------------------------------------
> 
> Dan Schaffer, Research Associate            e-mail: Daniel.S.Schaffer at noaa.gov
> NOAA/OAR/Earth System Research Lab         phone:  (303) 497-7252
> Aviation Branch                             fax:    (303) 497-6301
> R/GSD5   325 Broadway                       
> Boulder, CO 80305
> 
> www-ad.fsl.noaa.gov/ac/schaffer.html
> _______________________________________________
> Met_help mailing list
> Met_help at mailman.ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/met_help


More information about the Met_help mailing list