[Met_help] [rt.rap.ucar.edu #39786] History for point_stat grib versus >1 point observation
RAL HelpDesk {for John Halley Gotway}
met_help at ucar.edu
Wed Aug 4 12:50:10 MDT 2010
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
Hi,
I am not sure about how it works the point_stat tool when the obs_file
includes more than 1 point.
I've runned the application with a obs_file (enclosed file) with 2
observation points. But the results obtained in the *cnt.txt output
file only presents results for 1 point (4 lines= 2 varibles(TMP,RH) +
2 interpolation methods). I have verified that the mean OBS value
presented in each line corresponds to the mean of the first obs point
for the selected time (including the values of the time window defined
with beg_ds and end_ds).
My question is: Do I have to run the point_stat for each observation
point and each time (every 6h for instance)?
If I have obs values every 10minutes do you recommend me some time
window (beg_ds,end_ds)?
Thank you very much,
Laura
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point observation
From: John Halley Gotway
Time: Tue Aug 03 09:53:39 2010
Laura,
You definitely do not need to run it once per observation value!
Generally speaking, here's how running point-stat works...
(1) Your forecast contains gridded model data, preferably in GRIB
version 1 format.
(2) You gather point observations from whatever source you'd like. If
you're using PREPBUFR point observations, you run them through PB2NC.
If you're using ASCII point observations, you format them
as needed, and run them through ASCII2NC.
(3) Next you set up the Point-Stat configuration file. In there you
specify things like which variables/levels you'd like to verify, the
observation message type(s) you'd like to use, the time window
for matching observations to your forecast, the threshold(s) you want
to apply to each field, and the areas over which you'd like to
accumulate statistics.
In specifying the areas over which to verify, you can specify it as
the full domain of your model...
mask_grid[] = [ "FULL" ];
Or as a subset of your domain by defining a lat/lon polyline and
specifying it using the "mask_poly" option.
There's probably a good reason why one of your points was included and
the other was not. I do hope that you'll ultimately be able to use
more than 2 points in your analysis because statistics
computed over two points won't be very meaningful. Although, if you
run point_stat for many valid times, you could use the stat_analysis
tool to aggregate the statistics at each location through time.
Please do the following:
(1) If you haven't already downloaded and installed the latest patches
for METv2.0, please do so from here:
http://www.dtcenter.org/met/users/support/known_issues/METv2.0/index.php
(2) Please send me the files you're using to run point_stat,
specifically your input forecast GRIB file, your input observations
(NetCDF and ASCII if you have both), and the configuration file you're
using. You can post them to our anonymous ftp site as follows:
ftp ftp.rap.ucar.edu
username=anonymous
password="your email address"
cd incoming/irap/met_help/trapero_data
put "each one of your input files"
bye
Please write me back to let me know when the data's there. I'll run
it here and try to figure out the reason why the second point isn't
being included.
Thanks,
John Halley Gotway
met_help at ucar.edu
RAL HelpDesk {for ltrapero.cenma} wrote:
> Tue Aug 03 06:48:10 2010: Request 39786 was acted upon.
> Transaction: Ticket created by ltrapero.cenma at iea.ad
> Queue: met_help
> Subject: point_stat grib versus >1 point observation
> Owner: Nobody
> Requestors: ltrapero.cenma at iea.ad
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=39786 >
>
>
> Hi,
>
> I am not sure about how it works the point_stat tool when the
obs_file
> includes more than 1 point.
>
> I've runned the application with a obs_file (enclosed file) with 2
> observation points. But the results obtained in the *cnt.txt output
> file only presents results for 1 point (4 lines= 2 varibles(TMP,RH)
+
> 2 interpolation methods). I have verified that the mean OBS value
> presented in each line corresponds to the mean of the first obs
point
> for the selected time (including the values of the time window
defined
> with beg_ds and end_ds).
>
> My question is: Do I have to run the point_stat for each observation
> point and each time (every 6h for instance)?
> If I have obs values every 10minutes do you recommend me some time
> window (beg_ds,end_ds)?
>
> Thank you very much,
>
> Laura
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point observation
From: ltrapero.cenma
Time: Wed Aug 04 02:51:30 2010
Thank you for clarifying this point.
Now I know that the problem was the area defined with mask_grid and
mask_poly options.
My main domain has 1km resolution over an area of 500km2 with complex
terrain (Pyrenees Mountains between Spain and France). In this area I
have 6
meteorological stations and I would like to validate the model for
selected
events (1-2days). I wanted to calculated independent statistics for
each
observation point in order to observe different behaviours depending
on the
elevation and the northerly location (=greater exposition to northerly
flows; snow episodes).
I have model outputs every hour and the observations every 10minuts;
Then my
beg_ds, end_ds can be 0seconds or may I have to consider a time
window?
[2days * 24h = 48 values for each point to calculate statistics]
Thank you very much
Laura
P Realment necessites imprimir aquest correu? El medi ambient és cosa
de
tots!
-----Mensaje original-----
De: RAL HelpDesk {for John Halley Gotway} [mailto:met_help at ucar.edu]
Enviado el: martes, 03 de agosto de 2010 17:54
Para: ltrapero.cenma at iea.ad
Asunto: Re: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point
observation
Laura,
You definitely do not need to run it once per observation value!
Generally
speaking, here's how running point-stat works...
(1) Your forecast contains gridded model data, preferably in GRIB
version 1
format.
(2) You gather point observations from whatever source you'd like. If
you're using PREPBUFR point observations, you run them through PB2NC.
If
you're using ASCII point observations, you format them
as needed, and run them through ASCII2NC.
(3) Next you set up the Point-Stat configuration file. In there you
specify
things like which variables/levels you'd like to verify, the
observation
message type(s) you'd like to use, the time window
for matching observations to your forecast, the threshold(s) you want
to
apply to each field, and the areas over which you'd like to accumulate
statistics.
In specifying the areas over which to verify, you can specify it as
the full
domain of your model...
mask_grid[] = [ "FULL" ];
Or as a subset of your domain by defining a lat/lon polyline and
specifying
it using the "mask_poly" option.
There's probably a good reason why one of your points was included and
the
other was not. I do hope that you'll ultimately be able to use more
than 2
points in your analysis because statistics
computed over two points won't be very meaningful. Although, if you
run
point_stat for many valid times, you could use the stat_analysis tool
to
aggregate the statistics at each location through time.
Please do the following:
(1) If you haven't already downloaded and installed the latest patches
for
METv2.0, please do so from here:
http://www.dtcenter.org/met/users/support/known_issues/METv2.0/index.php
(2) Please send me the files you're using to run point_stat,
specifically
your input forecast GRIB file, your input observations (NetCDF and
ASCII if
you have both), and the configuration file you're
using. You can post them to our anonymous ftp site as follows:
ftp ftp.rap.ucar.edu
username=anonymous
password="your email address"
cd incoming/irap/met_help/trapero_data
put "each one of your input files"
bye
Please write me back to let me know when the data's there. I'll run
it here
and try to figure out the reason why the second point isn't being
included.
Thanks,
John Halley Gotway
met_help at ucar.edu
RAL HelpDesk {for ltrapero.cenma} wrote:
> Tue Aug 03 06:48:10 2010: Request 39786 was acted upon.
> Transaction: Ticket created by ltrapero.cenma at iea.ad
> Queue: met_help
> Subject: point_stat grib versus >1 point observation
> Owner: Nobody
> Requestors: ltrapero.cenma at iea.ad
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=39786 >
>
>
> Hi,
>
> I am not sure about how it works the point_stat tool when the
obs_file
> includes more than 1 point.
>
> I've runned the application with a obs_file (enclosed file) with 2
> observation points. But the results obtained in the *cnt.txt output
> file only presents results for 1 point (4 lines= 2 varibles(TMP,RH)
+
> 2 interpolation methods). I have verified that the mean OBS value
> presented in each line corresponds to the mean of the first obs
point
> for the selected time (including the values of the time window
defined
> with beg_ds and end_ds).
>
> My question is: Do I have to run the point_stat for each observation
> point and each time (every 6h for instance)?
> If I have obs values every 10minutes do you recommend me some time
> window (beg_ds,end_ds)?
>
> Thank you very much,
>
> Laura
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point observation
From: ltrapero.cenma
Time: Wed Aug 04 07:56:22 2010
HI,
Thank you for clarifying this point.
Now I am sure that the problem was the area defined with mask_grid and
mask_poly options.
My main domain has 1km resolution over an area (500km2) with complex
terrain (Pyrenees Mountains between Spain and France). In this area I
have 6
meteorological stations and I would like to validate the model for
selected
events (1-2days). I wanted to calculated independent statistics for
each
observation point in order to observe different behaviors depending on
the
elevation and its northerly location (=greater exposition to northerly
flows,specially in snow episodes).
[2days * 24h = 48 values for each point to calculate statistics]
I have model outputs every hour and the observations every 10minuts;
Then my
beg_ds, end_ds could it be 0 seconds or may I have to consider a time
window?
Now I amb trying to run the WPPV· to transform the wrfout* files to
GRIB 1 files.I have problems with when I execute wrfpost.exe.
Because of these problems I have tested the point_stat application
with the FNL grib files available on internet.
Thank you very much
Laura
P Realment necessites imprimir aquest correu? El medi ambient és
cosa de
tots!
-----Mensaje original-----
De: RAL HelpDesk {for John Halley Gotway} [mailto:met_help at ucar.edu]
Enviado el: martes, 03 de agosto de 2010 17:54
Para: ltrapero.cenma at iea.ad
Asunto: Re: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point
observation
Laura,
You definitely do not need to run it once per observation value!
Generally
speaking, here's how running point-stat works...
(1) Your forecast contains gridded model data, preferably in GRIB
version 1
format.
(2) You gather point observations from whatever source you'd like. If
you're using PREPBUFR point observations, you run them through PB2NC.
If
you're using ASCII point observations, you format them
as needed, and run them through ASCII2NC.
(3) Next you set up the Point-Stat configuration file. In there you
specify
things like which variables/levels you'd like to verify, the
observation
message type(s) you'd like to use, the time window
for matching observations to your forecast, the threshold(s) you want
to
apply to each field, and the areas over which you'd like to accumulate
statistics.
In specifying the areas over which to verify, you can specify it as
the full
domain of your model...
mask_grid[] = [ "FULL" ];
Or as a subset of your domain by defining a lat/lon polyline and
specifying
it using the "mask_poly" option.
There's probably a good reason why one of your points was included and
the
other was not. I do hope that you'll ultimately be able to use more
than 2
points in your analysis because statistics
computed over two points won't be very meaningful. Although, if you
run
point_stat for many valid times, you could use the stat_analysis tool
to
aggregate the statistics at each location through time.
Please do the following:
(1) If you haven't already downloaded and installed the latest patches
for
METv2.0, please do so from here:
http://www.dtcenter.org/met/users/support/known_issues/METv2.0/index.php
(2) Please send me the files you're using to run point_stat,
specifically
your input forecast GRIB file, your input observations (NetCDF and
ASCII if
you have both), and the configuration file you're
using. You can post them to our anonymous ftp site as follows:
ftp ftp.rap.ucar.edu
username=anonymous
password="your email address"
cd incoming/irap/met_help/trapero_data
put "each one of your input files"
bye
Please write me back to let me know when the data's there. I'll run
it here
and try to figure out the reason why the second point isn't being
included.
Thanks,
John Halley Gotway
met_help at ucar.edu
RAL HelpDesk {for ltrapero.cenma} wrote:
> Tue Aug 03 06:48:10 2010: Request 39786 was acted upon.
> Transaction: Ticket created by ltrapero.cenma at iea.ad
> Queue: met_help
> Subject: point_stat grib versus >1 point observation
> Owner: Nobody
> Requestors: ltrapero.cenma at iea.ad
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=39786 >
>
>
> Hi,
>
> I am not sure about how it works the point_stat tool when the
> obs_file includes more than 1 point.
>
> I've runned the application with a obs_file (enclosed file) with 2
> observation points. But the results obtained in the *cnt.txt output
> file only presents results for 1 point (4 lines= 2 varibles(TMP,RH)
> + 2 interpolation methods). I have verified that the mean OBS value
> presented in each line corresponds to the mean of the first obs
> point for the selected time (including the values of the time
> window defined with beg_ds and end_ds).
>
> My question is: Do I have to run the point_stat for each observation
> point and each time (every 6h for instance)?
> If I have obs values every 10minutes do you recommend me some time
> window (beg_ds,end_ds)?
>
> Thank you very much,
>
> Laura
>
>
----- Fi del missatge reenviat -----
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point observation
From: John Halley Gotway
Time: Wed Aug 04 09:35:55 2010
Laura,
That's a very good question about beg_ds and end_ds being set to 0.
Yes, I do believe that will work. The time window is defined as
[fcst_time + beg_ds, fcst_time + end_ds]. If they're both set to
0, then Point-Stat will only use observations that are valid at
exactly the same time as the forecast (down to the second).
It sounds like you're actually interested in verifying through time at
each observation location. So each time you run Point-Stat, every
hour during your 2-day time period, all you really care about
is getting the matched pair values (forecast and observation value for
each observation location) for that time. And then you'd like to
compute statistics through time using those matched pairs at
each station (48 pairs for each station over 2 days). If this is the
situation you're in, here's what I'd suggest doing...
(1) Run Point-Stat on your model output every hour, but configure
Point-Stat with the following settings:
// To only use observations valid at your exact forecast valid time
beg_ds = 0;
end_ds = 0;
// Don't compute statistics using an aerial average during each
run.
// Instead, just look at the values at each observation location.
mask_grid[] = [];
mask_poly[] = [];
mask_sid = "station.list"; // Where station.list is an ASCII file
containing a whitespace-separated list of you 6 station names
// Only output the matched-pair line type for each run of Point-
Stat
output_flag[] = [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1 ];
You're welcome to set "interp_method" and "interp_width" however you'd
like to choose the interpolation method(s) you'd like to use.
The following choices would use the nearest neighbor and bilinear
interpolation methods:
interp_method[] = [ "UW_MEAN", "DW_MEAN" ];
interp_width[] = [ 1, 2 ];
Running Point-Stat in this way will have the effect of dumping the
matched pair value at each station into the output. There will be one
matched pair (MPR) line in the output for each combination of
forecast variable/interpolation method/station name. That line will
contain the observation value and the forecast value interpolated to
the observation location.
(2) After you've run Point-Stat for each hour, use the STAT-Analysis
tool to aggregate together your results over the two day period. Try
running a job like this:
stat_analysis -lookin /directory/containing/point-stat/output \
-job aggregate_stat -line_type MPR -out_line_type CNT -v 2 \
-fcst_var TMP -vx_mask STATION_NAME -interp_mthd DW_MEAN
-interp_pnts 4 \
-dump_row aggr_job.stat
This will parse the STAT output files from the Point-Stat runs, filter
out only those MPR lines for TMP (or whatever weather variable you're
verifying) using the interpolation method DW_MEAN/4 at the
station named "STATION_NAME". In your case, this job should find 48
lines of STAT data - one for each hour. It'll read those matched
pairs and compute continuous statistics for them. I suggest
using the "-dump_row" option to make sure the job is running over the
exact subset of STAT data lines that you intend. You'd run this job
six times, once for each station in your dataset.
You can also set "-out_line_type" equal to other line types, like CTC
for contingency table counts or CTS for contingency table statistics,
but those will require additional arguments - like which
thresholds should be used.
Hopefully, this process will help you compute the statistics you're
after. Also, with such a small number of matched, you could consider
using the R statistical programming language to read the MPR
output from Point-Stat for doing more data analysis.
Thanks,
John
RAL HelpDesk {for ltrapero.cenma} wrote:
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=39786 >
>
> HI,
> Thank you for clarifying this point.
> Now I am sure that the problem was the area defined with mask_grid
and
> mask_poly options.
>
> My main domain has 1km resolution over an area (500km2) with complex
> terrain (Pyrenees Mountains between Spain and France). In this area
I have 6
> meteorological stations and I would like to validate the model for
selected
> events (1-2days). I wanted to calculated independent statistics for
each
> observation point in order to observe different behaviors depending
on the
> elevation and its northerly location (=greater exposition to
northerly
> flows,specially in snow episodes).
> [2days * 24h = 48 values for each point to calculate statistics]
>
> I have model outputs every hour and the observations every 10minuts;
Then my
> beg_ds, end_ds could it be 0 seconds or may I have to consider a
time window?
>
> Now I amb trying to run the WPPV· to transform the wrfout* files to
> GRIB 1 files.I have problems with when I execute wrfpost.exe.
> Because of these problems I have tested the point_stat application
> with the FNL grib files available on internet.
>
>
> Thank you very much
>
> Laura
>
> P Realment necessites imprimir aquest correu? El medi ambient és
cosa de
> tots!
> -----Mensaje original-----
> De: RAL HelpDesk {for John Halley Gotway} [mailto:met_help at ucar.edu]
> Enviado el: martes, 03 de agosto de 2010 17:54
> Para: ltrapero.cenma at iea.ad
> Asunto: Re: [rt.rap.ucar.edu #39786] point_stat grib versus >1 point
> observation
>
> Laura,
>
> You definitely do not need to run it once per observation value!
Generally
> speaking, here's how running point-stat works...
> (1) Your forecast contains gridded model data, preferably in GRIB
version 1
> format.
> (2) You gather point observations from whatever source you'd like.
If
> you're using PREPBUFR point observations, you run them through
PB2NC. If
> you're using ASCII point observations, you format them
> as needed, and run them through ASCII2NC.
> (3) Next you set up the Point-Stat configuration file. In there you
specify
> things like which variables/levels you'd like to verify, the
observation
> message type(s) you'd like to use, the time window
> for matching observations to your forecast, the threshold(s) you
want to
> apply to each field, and the areas over which you'd like to
accumulate
> statistics.
>
> In specifying the areas over which to verify, you can specify it as
the full
> domain of your model...
> mask_grid[] = [ "FULL" ];
> Or as a subset of your domain by defining a lat/lon polyline and
specifying
> it using the "mask_poly" option.
>
> There's probably a good reason why one of your points was included
and the
> other was not. I do hope that you'll ultimately be able to use more
than 2
> points in your analysis because statistics
> computed over two points won't be very meaningful. Although, if you
run
> point_stat for many valid times, you could use the stat_analysis
tool to
> aggregate the statistics at each location through time.
>
> Please do the following:
> (1) If you haven't already downloaded and installed the latest
patches for
> METv2.0, please do so from here:
>
http://www.dtcenter.org/met/users/support/known_issues/METv2.0/index.php
> (2) Please send me the files you're using to run point_stat,
specifically
> your input forecast GRIB file, your input observations (NetCDF and
ASCII if
> you have both), and the configuration file you're
> using. You can post them to our anonymous ftp site as follows:
>
> ftp ftp.rap.ucar.edu
> username=anonymous
> password="your email address"
> cd incoming/irap/met_help/trapero_data
> put "each one of your input files"
> bye
>
> Please write me back to let me know when the data's there. I'll run
it here
> and try to figure out the reason why the second point isn't being
included.
>
> Thanks,
> John Halley Gotway
> met_help at ucar.edu
>
>
>
> RAL HelpDesk {for ltrapero.cenma} wrote:
>> Tue Aug 03 06:48:10 2010: Request 39786 was acted upon.
>> Transaction: Ticket created by ltrapero.cenma at iea.ad
>> Queue: met_help
>> Subject: point_stat grib versus >1 point observation
>> Owner: Nobody
>> Requestors: ltrapero.cenma at iea.ad
>> Status: new
>> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=39786 >
>>
>>
>> Hi,
>>
>> I am not sure about how it works the point_stat tool when the
>> obs_file includes more than 1 point.
>>
>> I've runned the application with a obs_file (enclosed file) with 2
>> observation points. But the results obtained in the *cnt.txt output
>> file only presents results for 1 point (4 lines= 2 varibles(TMP,RH)
>> + 2 interpolation methods). I have verified that the mean OBS
value
>> presented in each line corresponds to the mean of the first obs
>> point for the selected time (including the values of the time
>> window defined with beg_ds and end_ds).
>>
>> My question is: Do I have to run the point_stat for each
observation
>> point and each time (every 6h for instance)?
>> If I have obs values every 10minutes do you recommend me some time
>> window (beg_ds,end_ds)?
>>
>> Thank you very much,
>>
>> Laura
>>
>>
>
>
>
>
>
> ----- Fi del missatge reenviat -----
>
>
>
------------------------------------------------
More information about the Met_help
mailing list