[Met_help] [rt.rap.ucar.edu #42124] History for question about point stat

[RAL HelpDesk {for John Halley Gotway}]

----------------------------------------------------------------
  Initial Request
----------------------------------------------------------------


John,

I am in the process of running point-stat to obtain statistics for
individual stations.  Point stat is running just fine, but I am definitely
at a loss because my RMSE values are off the charts.

Could this be because I used individual stations (mask_sid) in the
configuration file with the "FULL" masking grid (instead of specifying a
list of individual stations)?

Let me know,
James

----------------------------------------------------------------
  Complete Ticket History
----------------------------------------------------------------

Subject: Re: [rt.rap.ucar.edu #42124] question about point stat
From: John Halley Gotway
Time: Mon Nov 15 12:01:38 2010

James,

When you supply a list of stations using the "mask_sid" option in
Point-Stat, each of those stations is treated as a separate masking
region.  Since you only run Point-Stat once for each time, you're
likely only getting 1 matched pair at each station.  An RMSE value
computed using 1 matched pair is useless.

Really, what you'd probably like to do is see how the model performs
at individual stations through time.  One way to do this would be to
do the following:

(1) Run Point-Stat once per verification time, and dump out the
matched pair (MPR) line type.
   - If you're only interested in model performance at individual
stations and NOT doing any areal average, just could provide a list of
the stations in which you're interested using the "mask_sid"
option.  If you do this, then I'd suggest ONLY writing out the MPR
line type to get the matched pairs.
   - If you are interested in some areal averages, you could specify
the masking regions using the "mask_grid" or "mask_poly" options.  Do
not use the "mask_sid" option.  Then in your output you'll
see RMSE values averaged over the areas you used, and you'll have the
MPR data for each matched pair that fell in those regions.

(2) Once you have Point-Stat output for several times, you could run
the STAT-Analysis data to recompute statistics for a single station
through time.  You'd run "aggregate_stat" type jobs to read MPR
lines in and write out various types of statistics.

Really you have some options here.  But exactly how you should
structure it really depends on the types of verification tasks you
want to perform.

Thanks,
John

On 11/12/2010 04:11 PM, RAL HelpDesk {for James P Cipriani} wrote:
>
> Fri Nov 12 16:11:20 2010: Request 42124 was acted upon.
> Transaction: Ticket created by jpcipria at us.ibm.com
>        Queue: met_help
>      Subject: question about point stat
>        Owner: Nobody
>   Requestors: jpcipria at us.ibm.com
>       Status: new
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=42124 >
>
>
>
> John,
>
> I am in the process of running point-stat to obtain statistics for
> individual stations.  Point stat is running just fine, but I am
definitely
> at a loss because my RMSE values are off the charts.
>
> Could this be because I used individual stations (mask_sid) in the
> configuration file with the "FULL" masking grid (instead of
specifying a
> list of individual stations)?
>
> Let me know,
> James

------------------------------------------------