[Met_help] [rt.rap.ucar.edu #66522] History for Aggregating Matched Pairs

[John Halley Gotway via RT]

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

Hello MET,

I am running an operational verification system that generates daily point stat files consisting of matched pairs.  In order to conserve memory on our servers, the models from which the daily PS files are generated can only be kept for a few days, and the point stat files themselves can only be kept for a week.

At the end of the week, I would like to make time average of skill scores (RMSE, MAE, etc.) over the week at each individual matched pair using the point-stat files that I have already generated throughout the week (I am doing this in order to have some kind of visual quality control for our observation stations).

Now, I know that the time-series tool is set up to do exactly this, but that would require a week worth of model data (which I do not have), and would also waste the MPR point-stat files which I am already saving throughout the week.  So, my question is:  Is there a way to iterate through matched pair point-stat files with the stat_aggregate tool, which will output matched pairs which have been aggregated over time (instead of by case, fcst_hour, parameter, etc.).  I assume there is a way to do this, but I have not able to think of it yet.  Please send me a message when you have a chance.

Thank you in advance

- Andrew


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

Subject: Re: [rt.rap.ucar.edu #66522] Aggregating Matched Pairs
From: John Halley Gotway
Time: Tue Apr 29 11:45:15 2014

Hello Andrew,

I see that your running Point-Stat and generating matched pair output
lines.  Due to storage limitations, you can't store more than a week
of Point-Stat output.  At the end of each week, you'd like to
compute some summary statistics for each station.  Is that all
correct?

You mentioned the series-analysis tool, but that actually is not what
you want here.  Series-Analysis compares gridded forecast data to
gridded observations.  Since you're running Point-Stat, you're
using point observations.

Instead, I'd suggest running the STAT-Analysis tool to compute partial
sums (SL1L2 line) and/or continuous statistics (CNT line) for each
station.  Here's an example...
    stat_analysis -lookin /path/to/point_stat/output \
       -job aggregate_stat -line_type MPR -out_line_type CNT \
       -fcst_var TMP -fcst_lev Z2 \
       -fcst_init_beg YYYYMMDD_HH -fcst_init_end YYYYMMDD_HH \
       -by OBS_SID \

This job will...
  (1) Read the MPR lines from the .stat files it finds in the
directory you pass to it
  (2) Filter out only those lines whose model initialization time
falls in the window you define with -fcst_init_beg and -fcst_init_end
  (3) Filter out only those lines for 2-m temperature
  (3) For each unique station id name (OBS_SID), it'll compute
continuous statistics for all of the lines it finds

Is this the type of functionality you're looking for?  You could run
similar jobs for other combinations of forecast variable and level.
STAT-Analysis has a lot of flexibility in defining the
filtering criteria - so you could play around with it to get it to do
exactly what you want.

Once you've defined the exact type of job(s) you want to run on the
command line, I'd suggest grouping them together in the configuration
file.  Then you can just run it at the end of each week before
scrubbing your data.

Hope that helps get you going.  If you have more specific questions
about how to get STAT-Analysis to do exactly what you want, please let
me know.

Thanks,
John Halley Gotway
met_help at ucar.edu

On 04/29/2014 09:09 AM, Andrew J. via RT wrote:
>
> Tue Apr 29 09:09:27 2014: Request 66522 was acted upon.
> Transaction: Ticket created by andrewwx at yahoo.com
>         Queue: met_help
>       Subject: Aggregating Matched Pairs
>         Owner: Nobody
>    Requestors: andrewwx at yahoo.com
>        Status: new
>   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=66522 >
>
>
> Hello MET,
>
> I am running an operational verification system that generates daily
point stat files consisting of matched pairs.  In order to conserve
memory on our servers, the models from which the daily PS files are
generated can only be kept for a few days, and the point stat files
themselves can only be kept for a week.
>
> At the end of the week, I would like to make time average of skill
scores (RMSE, MAE, etc.) over the week at each individual matched pair
using the point-stat files that I have already generated throughout
the week (I am doing this in order to have some kind of visual quality
control for our observation stations).
>
> Now, I know that the time-series tool is set up to do exactly this,
but that would require a week worth of model data (which I do not
have), and would also waste the MPR point-stat files which I am
already saving throughout the week.  So, my question is:  Is there a
way to iterate through matched pair point-stat files with the
stat_aggregate tool, which will output matched pairs which have been
aggregated over time (instead of by case, fcst_hour, parameter, etc.).
I assume there is a way to do this, but I have not able to think of it
yet.  Please send me a message when you have a chance.
>
> Thank you in advance
>
> - Andrew
>

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