[Met_help] Masking Regions

John Halley Gotway johnhg at rap.ucar.edu
Tue Mar 16 14:52:31 MDT 2010


Since you're doing gridded verification over a large domain, I'd strongly
suggest disabling the bootstrap confidence intervals.  In fact, we should
probably make that the default in the Grid-Stat configuration file anyway.
 So in the config file, try setting:
n_boot_rep = 0;

When you rerun, you should see a very dramatic improvement in the runtime.
 In fact, I'm guessing it'll run as fast as you need it to at that point.

Also, I'll briefly address your other questions...

- Both mask_poly and mask_grid are used to define "masking regions" to
define the area of which to perform the verification.  "mask_grid" can be
used to specify one of the pre-defined NCEP verification grids. 
"mask_poly" is more useful in that you can define your own masking region
as a lat/lon polyline.  To make things run as fast as possible, choose the
masking region you'd like to use and run it through the gen-poly-mask
tool.  Then use the output of gen-poly-mask in the "mask_poly" entry of
the Grid-Stat configuration file.

- The predefined NCEP grids are described here:


> John, I am now getting things to run and generate output.  Now the issue
> is
> finding ways to speed up the verification.  I am currently verifying the
> full stage IV grid and it takes over three hours per each days model run.
> Attached is the configuration files I am using.  I thought about reducing
> the areal coverage to speed up the process.  What is the difference
> between
> mask_poly and mask_grid?  Where can I find a description of the NCEP grid
> codes?  I used mask_poly = east.poly to limit the verification to the east
> half of the US.   I am interpreting this correctly?  This verify has been
> running an hour just for one day - at this rate I will never get done in a
> reasonable time.  What could I change in my config file to get a quicker
> output?  I am running this on our development system - ( power 4 plus, ibm
> p655 1.5 gigahertz -8 way nodes )
> Thanks
> Bob

More information about the Met_help mailing list