[Met_help] [rt.rap.ucar.edu #88657] History for Bad Data File

[John Halley Gotway via RT]

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

John, I attached the file you need.



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

Subject: Bad Data File
From: John Halley Gotway
Time: Fri Jan 25 20:59:29 2019

Bob,

Great, I got the other attachment you sent.  I still don't know
exactly
what Stat-Analysis job you're running, but I made an educated guess
which
did result in a runtime error:

*/usr/local/met-6.1/bin/stat_analysis \*
* -lookin point_stat_global_17km_all_120000L_20171205_120000V.stat \*
* -job aggregate_stat -line_type MPR -out_line_type PCT \*
*-out_fcst_thresh eq0.25 -out_obs_thresh gt0*

*ERROR  : Nx2ContingencyTable::inc_event(double) -> bad value ...
5.07874*

The MPR output from Point-Stat contains both the raw fcst/obs matched
pair
values and the neighborhood fractional coverage values.  When doing
probabilistic vx with an Nx2 contingency table, the forecast values
should
be probabilities between 0 and 1... not the raw APCP values, like
5.07874.

We can control this by telling Stat-Analysis to only use MPR lines
where
INTERP_MTHD = NBRHD_SQUARE.  Looks like your data contains output for
4
different thresholds (>1, >=10, >=25, >=50) and each of those
thresholds
requires a different job.  So I put multiple jobs into one config
file.

*/usr/local/met-6.1/bin/stat_analysis \*
*-lookin point_stat_global_17km_all_120000L_20171205_120000V.stat \*
*-config STATAnalysisConfig_hira*

And that generates these files:
*APCP_gt1.stat*


*APCP_ge10.statAPCP_ge25.statAPCP_ge50.stat*


Since you're running HiRA, I expected to see a larger value for
INTERP_PNTS.  Since you're computing HiRA in a neighborhood of size
one,
the probability values are always either 0 (no-event) or 1 (event).
Looking at your MPR data, that's exactly what I see... 0's and 1's in
the
FCST column.

So I'd recommend increasing the HiRA neighborhood size to larger
values...
which you were probably planning to do anyway.  That'll generate
smoother
probability values between 0 and 1 and produce more interesting output
from
STAT-Analysis.

In this config file, I'm dumping the PCT counts, but you can easily
change
the output line type to PSTD.

Thanks,
John

On Fri, Jan 25, 2019 at 3:36 PM The RT System itself via RT <
met_help at ucar.edu> wrote:

>
> Fri Jan 25 15:35:47 2019: Request 88657 was acted upon.
> Transaction: Given to johnhg (John Halley Gotway) by RT_System
>        Queue: met_help
>      Subject: Bad Data File
>        Owner: johnhg
>   Requestors: robert.craig.2 at us.af.mil
>       Status: deleted
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=88657 >
>
>
> This transaction appears to have no content
>

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