[Met_help] [rt.rap.ucar.edu #73371] History for obs/model plots with MET

[John Halley Gotway via RT]

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

Hello,
I looked at the presentation below and wonder if this kind of plots are available in MET 5.0? If not can you make 5.1 available for testing?
Thanks,
Mariusz




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

Subject: obs/model plots with MET
From: John Halley Gotway
Time: Thu Sep 10 16:56:09 2015

Mariusz,

That sort of analysis is available using MET version 5.0.  It's
generated
running a combination of the Point-Stat and Stat-Analysis tools, and
the
plot is created using NCL.

When you run Point-Stat, you'll need to dump out the MPR (matched
pair)
line type.  Unfortunately, the MPR output is a pretty poor way of
storing
data... you get one output line for each matched forecast/observation
pair
for each verification task.  You run Point-Stat many times... once for
each
model output time.

Then you run the following sort of Stat-Analysis command:
   stat_analysis \
   -lookin /path/to/mpr/data \
   -job aggregate_stat -line_type MPR -out_line_type CNT \
   -fcst_var TMP -fcst_lev Z2 -fcst_lead 12 \ (or some other set of
filtering criteria)
   -by OBS_SID,OBS_LAT,OBS_LON \
   -out agg_stat_mpr_to_cnt.out

That will search the directory named /path/to/mpr/data for files
ending in
".stat".  It'll read MPR lines from those files and keep the ones
where
FCST_VAR = TMP, FCST_LEV = Z2, and FCST_LEAD = 12 hours.  Then it'll
group
together the matched pairs based on unique combinations of the OBS_SID
(i.e. station id), OBS_LAT, and OBS_LON columns.  For each unique
combination, it'll compute and write out a continuous statistics
output
line.

That gives you statistics at individual stations computed through
time.  I
included the OBS_LAT and OBS_LON in the "-by" option so that their
values
will get passed through to the output.  That way you'll know where to
plot
the statistics.

Here's a link to an example that actually does something slightly
different.  It just reads MPR lines for a single time and plots the
FCST -
OBS difference:
   ftp://ftp.rap.ucar.edu/incoming/irap/met_help/for_pagowski

Save all 3 files from that ftp site to some directory and run:
   ncl plot_me.ncl

If everything goes well, you'll see an output image named
"fcst_gt300_obs_gt300.png", but there will likely be issues with
paths.

This generates an image but is very clunky and hard-coded.  Once you
get
the -by case output from stat-analysis, you can plot the statistics by
station using whatever plotting tool you'd like.

Hope that helps point you in the right direction.

Thanks,
John

On Thu, Sep 10, 2015 at 4:38 PM, John Halley Gotway <johnhg at ucar.edu>
wrote:

> Mariusz,
>
> That sort of analysis is available using MET version 5.0.  It's
generated
> running a combination of the Point-Stat and Stat-Analysis tools, and
the
> plot is created using NCL.
>
> When you run Point-Stat, you'll need to dump out the MPR (matched
pair)
> line type.  Unfortunately, the MPR output is a pretty poor way of
storing
> data... you get one output line for each matched
forecast/observation pair
> for each verification task.  You run Point-Stat many times... once
for each
> model output time.
>
> Then you run the following sort of Stat-Analysis command:
>    stat_analysis \
>    -lookin /path/to/mpr/data \
>    -job aggregate_stat -line_type MPR -out_line_type CNT \
>    -fcst_var TMP -fcst_lev Z2 -fcst_lead 12 \ (or some other set of
> filtering criteria)
>    -by OBS_SID,OBS_LAT,OBS_LON \
>    -out agg_stat_mpr_to_cnt.out
>
> That will search the directory named /path/to/mpr/data for files
ending in
> ".stat".  It'll read MPR lines from those files and keep the ones
where
> FCST_VAR = TMP, FCST_LEV = Z2, and FCST_LEAD = 12 hours.  Then it'll
group
> together the matched pairs based on unique combinations of the
OBS_SID
> (i.e. station id), OBS_LAT, and OBS_LON columns.  For each unique
> combination, it'll compute and write out a continuous statistics
output
> line.
>
> That gives you statistics at individual stations computed through
time.  I
> included the OBS_LAT and OBS_LON in the "-by" option so that their
values
> will get passed through to the output.  That way you'll know where
to plot
> the statistics.
>
> I've attached an NCL script that actually does something slightly
> different.  It just reads MPR lines for a single time and plots the
FCST -
> OBS difference.  I've also attached 2 other files that this script
reads.
> Save all 3 files to some directory and run:
>    ncl plot_me.ncl
>
> If everything goes well, you'll see an output image named
> "fcst_gt300_obs_gt300.png", but there will likely be issues with
paths.
>
> This generates an image but is very clunky.  Once you get the -by
case
> output from stat-analysis, you can plot the statistics by station
using
> whatever plotting tool you'd like.
>
> Hope that helps point you in the right direction.
>
> Thanks,
> John
>
>
>
>
> On Thu, Sep 10, 2015 at 12:02 PM, Mariusz Pagowski via RT <
> met_help at ucar.edu> wrote:
>
>>
>> Thu Sep 10 12:02:06 2015: Request 73371 was acted upon.
>> Transaction: Ticket created by Mariusz.Pagowski at noaa.gov
>>        Queue: met_help
>>      Subject: obs/model plots with MET
>>        Owner: Nobody
>>   Requestors: Mariusz.Pagowski at noaa.gov
>>       Status: new
>>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=73371 >
>>
>>
>> Hello,
>> I looked at the presentation below and wonder if this kind of plots
are
>> available in MET 5.0? If not can you make 5.1 available for
testing?
>> Thanks,
>> Mariusz
>>
>>
>>
>>
>

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