[Met_help] [rt.rap.ucar.edu #76013] History for Simple outlier stats

[John Halley Gotway via RT]

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

Hello MET Help,

I want to compute simple outlier statistics, such as % of stations with temperature error > 5°C. Is there a way to do this?

Also, is there a way to identify particular outliers? I can see them on a matched-pair plot, for instance, but I’d like to know which rows they correspond to.

Thank you.

Richard Carpenter


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

Subject: Simple outlier stats
From: John Halley Gotway
Time: Tue Apr 26 15:13:03 2016

Hello Richard,

I apologize for not getting back to you sooner on your recent question
to
MET-Help.  Great question!  I have been thinking about this but cannot
think of an elegant way of producing this output.

However, MET can do the following...

   - Run Point-Stat to read a gridded forecast file, compare to point
observations, and compute matched pair (MPR) output lines.
   - Run those MPR output lines through the STAT-Analysis tool and
filter
them by applying thresholds to any of the data columns.
   - The MPR line includes columns for the forecast and observation
values... but not their difference.
   - So you could easily filter out MPR lines where the FCST and/or
OBS
value meet some threshold criteria, but not their difference!

There is another tool in MET named "tc_stat" which is somewhat similar
to
the "stat_analysis" tool.  tc_stat reads the matched hurricane track
output
from the tc_pairs tool, filters it down, and runs analysis jobs on it.
I
mention this because tc_stat *CAN* compute differences and absolute
values
of columns when reading data.

The functionality that's really needed here is to enhance
stat_analysis to
filter data by applying a threshold to the difference of two columns,
namely FCST - OBS.  But that doesn't currently exist.

If I were in your shoes, I'd write a little script to read the MPR
output
lines, compute the FCST - OBS difference, and report out the lines
which
meet the threshold criteria.

Hope that helps clarify.

Thanks,
John Halley Gotway
met_help at ucar.edu





On Thu, Apr 21, 2016 at 9:09 AM, Richard Carpenter via RT
<met_help at ucar.edu
> wrote:

>
> Thu Apr 21 09:09:20 2016: Request 76013 was acted upon.
> Transaction: Ticket created by rcarpenter at wdtinc.com
>        Queue: met_help
>      Subject: Simple outlier stats
>        Owner: Nobody
>   Requestors: rcarpenter at wdtinc.com
>       Status: new
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=76013 >
>
>
> Hello MET Help,
>
> I want to compute simple outlier statistics, such as % of stations
with
> temperature error > 5°C. Is there a way to do this?
>
> Also, is there a way to identify particular outliers? I can see them
on a
> matched-pair plot, for instance, but I’d like to know which rows
they
> correspond to.
>
> Thank you.
>
> Richard Carpenter
>
>

------------------------------------------------
Subject: Simple outlier stats
From: John Halley Gotway
Time: Wed Apr 27 14:21:24 2016

Richard,

FYI, I did add a development task for the next release of MET (5.2) to
port
the TC-Stat functionality over to the STAT-Analysis tool.  That will
enable
STAT-Analysis to filter data by defining a difference and/or absolute
value
for columns of data.  So you could use "-column_thresh FCST-OBS >=5"
and
you'd select all MPR lines where the error (i.e. FCST-OBS) it >= 5.

I'll go ahead and resolve this issue, but hopefully we'll be able to
include that new functionality in version 5.2.

Thanks,
John


On Tue, Apr 26, 2016 at 3:13 PM, The RT System itself via RT <
met_help at ucar.edu> wrote:

>
> Tue Apr 26 15:13:03 2016: Request 76013 was acted upon.
> Transaction: Given to johnhg (John Halley Gotway) by RT_System
>        Queue: met_help
>      Subject: Simple outlier stats
>        Owner: johnhg
>   Requestors: rcarpenter at wdtinc.com
>       Status: new
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=76013 >
>
>
> This transaction appears to have no content
>

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