[Met_help] Point-Stat Tool --- Point Observation Component Issue

John Halley Gotway johnhg at rap.ucar.edu
Fri Aug 7 07:32:57 MDT 2009


Waylon,

Interesting.  Stat-Analysis is failing to parse the STAT data correctly as you found.  The reason is the name you used for the MODEL parameter in the Point-Stat configuration file:
model = "WFO CRP WRF-ARW";

The embedded spaces get written out by Point-Stat correctly, but when STAT-Analysis reads the STAT data, it gets confused by the embedded spaces.  It expects columns to be delimited by white space.
In the STAT file you sent, I tried replacing "WFO CRP WRF-ARW" with "WFO_CRP_WRF-ARW", and STAT-Analysis worked fine.

I suppose we should add error checking to the tools to prevent users from embedding white space into the model name.  Glad it's an easy fix.

Let me offer a reminder... when you run STAT-Analysis to do your aggregation and recomputation of statistics, be sure to start off using the "-dump_row" option to tell you exactly which stat lines are
being used for the job.  And verify that that's really the set of data you want to be using.  It's a good idea to use the "-dump_row" option initially until you're confident that the job really is
doing what you expect.

Thanks,
John

Waylon.Collins wrote:
> John,
> 
> You don't need to create any additional tools; I decided to write a
> shell script to run Point-Stat for multiple cases and then run
> Stat_Analysis to aggregate the results, as you suggested.
> 
> However, I am having difficulty running Stat_Analysis. Using Point-Stat,
> I generated STAT files for 30 days in a row. Then, I ran Stat_Analysis.
> My objective is to use "aggregate_stat", yet I can't even get results
> from "filter". Attached includes the output from running Stat_Analysis
> using the configuration file with the "filter" option. Also attached is
> the corresponding configuration file. Note that Stat_Analysis processes
> my 30 STAT files, yet does not appear to read the lines. The file
> following the -dump_row option had a size of 0. The filtering was set in
> retrieve lines that exist in the STAT files, but to no avail. I also
> attached the configuration file using in my Point-Stat run and included
> one of the 30 STAT files.
> 
> Any idea what I am doing wrong?
> 
> Waylon
> 
> John Halley Gotway wrote:
>> Waylon,
>>
>> The Point-Stat tool was designed to be run at a single moment in
>> time.  In the example you gave, for time 20090417_0600 (17 April 2009
>> at 0600 UTC), you'd run Point-Stat once passing to it the
>> forecast file valid at that time and the set of observations you'd
>> like to compare to those observations.
>>
>> If you'd like to compute quarterly or monthly verification statistics,
>> you'll need to run Point-Stat once for each verification time, and
>> then use the Stat-Analysis tool to aggregate the results.
>>
>> It depends on how much data we're talking about here, but one option
>> would be to make use of the MPR (matched pair) line type.  When you
>> run Point-Stat, you have the option of dumping out the fcst and
>> obs raw matched pair values.  Then you could run a job in the
>> Stat-Analysis tool to read in all those matched pairs for a month, for
>> example, and compute whatever statistics you'd like from them.
>>
>> However, the MPR line type is pretty inefficient since all the header
>> data is written out for each matched pair.  So if space is an issue,
>> you could instead make use of the SL1L2 partial sums and CTC
>> contingency table counts line types from Point-Stat.  Then you could
>> run Stat-Analysis jobs to aggregate those lines over a month and
>> compute statistics from them.
>>
>> As for passing multiple NetCDF files to the Point-Stat tool, please
>> take a look at the usage statement for Point-Stat in the
>> documentation.  The "-ncfile" command line option can be used as many
>> times
>> as you like to pass additional NetCDF point observation files to
>> Point-Stat to be used.  So you could convert the hourly files and pass
>> many of them to Point-Stat on the command line using "-ncfile".
>>  Or I could probably modify the AWIPS2NC tool to convert multiple
>> input NetCDF files into a single output NetCDF file... but I'm pretty
>> busy this week and next.
>>
>> Hope that helps.
>>
>> John
>>
>> Waylon.Collins wrote:
>>  
>>> Hello John,
>>>
>>> Yes, I was successful this time. Thanks.
>>>
>>> The AWIPS netCDF observation files I have been sending you are for a
>>> single moment in time (e.g. 20090417_0600 for 17 April 2009 at 0600
>>> UTC). However, I will be generating either quarterly or monthly
>>> verification statistics. How can I run Point-Stat in order to process
>>> many netCDF files at once? Alternatively, I my plan is to install the
>>> NCO (netCDF Operators) software (http://nco.sourceforge.net) and plan to
>>> use the ncrcat (netCDF Record Concatenator), which is part of NCO, in
>>> order to concatenates all of the hourly netCDF files into one netCDF
>>> file before running your awips2nc program. Any other ideas?
>>>
>>> Waylon
>>>
>>> John Halley Gotway wrote:
>>>    
>>>> Waylon,
>>>>
>>>> Thanks for sending me all that data.  It made debugging the problem
>>>> very easy.  Sorry about this, but I had my counts off by 1.  I was
>>>> indexing the header id's from 1 to N, when I should have been
>>>> doing it from 0 to N-1.  This amounts to a 1 line change in the
>>>> "awips2nc.cc" file.  I've attached the updated version.  Please
>>>> replace METv2.0/src/awips2nc/awips2nc.cc with the attached version,
>>>> recompile, and try running AWIPS2NC and Point-Stat again.
>>>>
>>>> Thanks,
>>>> John
>>>>
>>>> Waylon.Collins wrote:
>>>>  
>>>>      
>>>>> John,
>>>>>
>>>>> After compiling using the PGI Makefile, and after installing
>>>>> METv2.0_patches_20090701.tar from the MET website, I ran awips2nc
>>>>> without error. However, when I ran the Point Stat Tool as a test, I
>>>>> encountered the attached error. I uploaded the associated files to the
>>>>> /incoming/irap/met_help/collins_data subdirectory you prepared for me
>>>>> earlier on ftp.rap.ucar.edu server. The file 20090417_0600 and
>>>>> 20090417_0600.out are the input and output to the awips2nc program,
>>>>> respectively. the *arw_wrfout* GRIB1 file is the forecast file. The
>>>>> PointStatConfig_ARW file is the configuration file for Point Stat.
>>>>>
>>>>> Waylon
>>>>>
>>>>> John Halley Gotway wrote:
>>>>>           
>>>>>> Waylon,
>>>>>>
>>>>>> The updated PGI Makefile is attached.  It includes additions in 4
>>>>>> spots to include the building of the AWIPS2NC tool.
>>>>>>
>>>>>> Hope it all goes well.
>>>>>>
>>>>>> John
>>>>>>
>>>>>> Waylon.Collins wrote:
>>>>>>  
>>>>>>               
>>>>>>> Hello John,
>>>>>>>
>>>>>>> The Fortran Compiler that I use is the from the Portland Group
>>>>>>> (PGI),
>>>>>>> although I do have gfortran installed on my computer (that runs the
>>>>>>> Redhat Enterprise Linux v4 O/S.) I created both PGI (16 June
>>>>>>> 2009) and
>>>>>>> gfortran (tonight) versions of both netCDF and libbufr.a and
>>>>>>> tried to
>>>>>>> install MET again tonight using your new Makefile_gnu file and the
>>>>>>> gfortran compiled versions of netCDF and BUFRLIB. I am having
>>>>>>> difficulty, although it appears to be minor. Nevertheless, could you
>>>>>>> prepare a Makefile_pgi file for the building of MET including
>>>>>>> awips2nc?
>>>>>>> Thanks again for your help.
>>>>>>>
>>>>>>> Waylon
>>>>>>>
>>>>>>> John Halley Gotway wrote:
>>>>>>>                      
>>>>>>>> Waylon,
>>>>>>>>
>>>>>>>> I went ahead and just took a shot at it.  After talking to a
>>>>>>>> meteorologist here, I assumed that both report types, METAR and
>>>>>>>> SPECI,
>>>>>>>> indicate surface observations.  I've attached some code for you to
>>>>>>>> try out.
>>>>>>>>
>>>>>>>> Please copy the attached file "METv2.0_awips2nc.20090714.tar.gz",
>>>>>>>> into
>>>>>>>> the top-level METv2.0 directory.  Then uncompress and untar it:
>>>>>>>> gunzip METv2.0_awips2nc.20090714.tar.gz
>>>>>>>> tar -xvf METv2.0_awips2nc.20090714.tar
>>>>>>>>
>>>>>>>> This will create a new directory "METv2.0/src/awips2nc".  Also, it
>>>>>>>> will overwrite the file "Makefile_gnu" with an updated version with
>>>>>>>> commands for building the awips2nc tool.  I wasn't sure which
>>>>>>>> compilers you used to build MET... if it was something other
>>>>>>>> than the
>>>>>>>> GNU compilers, just look at the Makefile changes I made, and make
>>>>>>>> similar ones for the version of the Makefile you're using.
>>>>>>>>
>>>>>>>> ... or let me know which compilers, and I can make the Makefile
>>>>>>>> changes for you.
>>>>>>>>
>>>>>>>> Then just rebuild MET by typing "make" in the METv2.0 directory.
>>>>>>>> That
>>>>>>>> should create the "METv2.0/bin/awips2nc" executable which you
>>>>>>>> can use
>>>>>>>> to reformat your AWIPS point observation file.  Type
>>>>>>>> "METv2.0/bin/awips2nc" to see the usage statement.  It's pretty
>>>>>>>> straight-forward.
>>>>>>>>
>>>>>>>> Try it out and let me know how it goes.  Right now it's assuming
>>>>>>>> all
>>>>>>>> of the point observations to be at the surface.  Just let me
>>>>>>>> know if
>>>>>>>> we need to modify the logic to handle a wider variety of data
>>>>>>>> types you may encounter.
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> John Halley Gotway
>>>>>>>> johnhg at ucar.edu
>>>>>>>>
>>>>>>>> John Halley Gotway wrote:
>>>>>>>>  
>>>>>>>>                            
>>>>>>>>> Waylon,
>>>>>>>>>
>>>>>>>>> Thanks for checking in.  I am working on it, but have been
>>>>>>>>> slowed by
>>>>>>>>> needing to get ready for a MET tutorial we're giving next week.
>>>>>>>>> I'll
>>>>>>>>> get to work on it though and will get in touch with you when
>>>>>>>>> I have something for you.  I hope to make significant progress
>>>>>>>>> this
>>>>>>>>> week.
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> John
>>>>>>>>>
>>>>>>>>> Waylon.Collins at noaa.gov wrote:
>>>>>>>>>                                     
>>>>>>>>>> Hello John,
>>>>>>>>>>
>>>>>>>>>> Just wondering if you made any progress developing software to
>>>>>>>>>> convert AWIPS METAR netCDF files to Point Stat tool compatible
>>>>>>>>>> netCDF files?
>>>>>>>>>>
>>>>>>>>>> Waylon
>>>>>>>>>> Waylon Collins (Senior Forecaster)
>>>>>>>>>> NOAA/National Weather Service
>>>>>>>>>> 300 Pinson Drive
>>>>>>>>>> Corpus Christi, TX 78406
>>>>>>>>>> (United States)
>>>>>>>>>> Telephone: (361) 289-0959   ----- Original Message -----
>>>>>>>>>> From: John Halley Gotway <johnhg at rap.ucar.edu>
>>>>>>>>>> Date: Thursday, June 25, 2009 1:43 pm
>>>>>>>>>> Subject: Re: [Met_help] Point-Stat Tool --- Point Observation
>>>>>>>>>> Component Issue
>>>>>>>>>> To: "Waylon.Collins" <Waylon.Collins at noaa.gov>, met_help
>>>>>>>>>> <met_help at ucar.edu>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>                                             
>>>>>>>>>>> Waylon,
>>>>>>>>>>>
>>>>>>>>>>> Sounds good.  I'm busy this week with the WRF-User's
>>>>>>>>>>> Workshop. But
>>>>>>>>>>> I'll take a look at these files next week.
>>>>>>>>>>>
>>>>>>>>>>> Thanks,
>>>>>>>>>>> John
>>>>>>>>>>>
>>>>>>>>>>> Waylon.Collins wrote:
>>>>>>>>>>>                                                     
>>>>>>>>>>>> Hello John,
>>>>>>>>>>>>
>>>>>>>>>>>> Yes, lets pursue the strategy of developing a reformatting tool
>>>>>>>>>>>> that
>>>>>>>>>>>> would read an AWIPS netCDF METAR file and write a Point-Stat
>>>>>>>>>>>> tool
>>>>>>>>>>>> compatible netCDF file. I uploaded three (3) METAR files to the
>>>>>>>>>>>> collins_data subdirectory on your FTP site. Go ahead and ignore
>>>>>>>>>>>> the                                          
>>>>>>>>>>>>                       
>>>>>>>>>>> QC
>>>>>>>>>>>                                                     
>>>>>>>>>>>> information and simply use the actual data values. The twelve
>>>>>>>>>>>> (12)
>>>>>>>>>>>> observation types you mentioned below are sufficient. Thanks
>>>>>>>>>>>> for
>>>>>>>>>>>>                                                                 
>>>>>>>>>>> your help.
>>>>>>>>>>>                                                     
>>>>>>>>>>>> Waylon
>>>>>>>>>>>>
>>>>>>>>>>>> John Halley Gotway wrote:
>>>>>>>>>>>>                                                             
>>>>>>>>>>>>> Waylon,
>>>>>>>>>>>>>
>>>>>>>>>>>>> We do not currently have tools that will do either of the
>>>>>>>>>>>>> things you
>>>>>>>>>>>>> requested.  However, I took at look at the NetCDF header you
>>>>>>>>>>>>> sent,                                                
>>>>>>>>>>>>>                         
>>>>>>>>>>> and
>>>>>>>>>>>                                                     
>>>>>>>>>>>>> the format looks pretty straight-forward.  I'd be happy to
>>>>>>>>>>>>> work with you on a reformatting tool that would read a NetCDF
>>>>>>>>>>>>> AWIPS
>>>>>>>>>>>>> file and write a NetCDF file for input into the Point-Stat
>>>>>>>>>>>>> tool.  I
>>>>>>>>>>>>> don't think it'd be too difficult.  But I'd need you to
>>>>>>>>>>>>> send me
>>>>>>>>>>>>> a few sample files for testing and answer some questions
>>>>>>>>>>>>> about the
>>>>>>>>>>>>> format.
>>>>>>>>>>>>>
>>>>>>>>>>>>> If you'd like to proceed with this, please send a few sample
>>>>>>>>>>>>> files
>>>>>>>>>>>>> either via email or by posting them on our anonymous FTP site:
>>>>>>>>>>>>> ftp ftp.rap.ucar.edu
>>>>>>>>>>>>> username = anonymous
>>>>>>>>>>>>> password = "your email address"
>>>>>>>>>>>>> cd incoming/irap/met_help/collins_data
>>>>>>>>>>>>> put "your sample files"
>>>>>>>>>>>>> bye
>>>>>>>>>>>>>
>>>>>>>>>>>>> Looking at the NetCDF header you sent, it looks like there's a
>>>>>>>>>>>>> lot                                                
>>>>>>>>>>>>>                         
>>>>>>>>>>> of
>>>>>>>>>>>                                                     
>>>>>>>>>>>>> quality control information in the file.  The goal of this
>>>>>>>>>>>>> reformatting tool would be to extract the actual observation
>>>>>>>>>>>>> values
>>>>>>>>>>>>> you'd like to use in the verification.  Would it be reasonable
>>>>>>>>>>>>> initially to ignore the QC information and just use the actual
>>>>>>>>>>>>> data
>>>>>>>>>>>>> values?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Here are the observation types I see that we would make
>>>>>>>>>>>>> sense to
>>>>>>>>>>>>> extract to use in verification.  Each is followed by the GRIB
>>>>>>>>>>>>> code                                                
>>>>>>>>>>>>>                         
>>>>>>>>>>> to
>>>>>>>>>>>                                                     
>>>>>>>>>>>>> which it would correspond:
>>>>>>>>>>>>> (1)  Sea Level Pressure (GRIB Code 2)
>>>>>>>>>>>>> (2)  Temperature (GRIB Code 11)
>>>>>>>>>>>>> (3)  Maximum Temperature (GRIB Code 15)
>>>>>>>>>>>>> (4)  Minimum Temperature (GRIB Code 16)
>>>>>>>>>>>>> (5)  Dewpoint (GRIB Code 17)
>>>>>>>>>>>>> (6)  Wind Direction (GRIB Code 31)
>>>>>>>>>>>>> (7)  Wind Speed (GRIB Code 32)
>>>>>>>>>>>>> (8)  Wind Gust (GRIB Code 180)
>>>>>>>>>>>>> (9)  1-hourly precip (GRIB Code 61)
>>>>>>>>>>>>> (10) 3-hourly precip (GRIB Code 61)
>>>>>>>>>>>>> (11) 6-hourly precip (GRIB Code 61)
>>>>>>>>>>>>> (12) 24-hourly precip (GRIB Code 61)
>>>>>>>>>>>>>
>>>>>>>>>>>>> Does that set of observation types look reasonable to you?  Or
>>>>>>>>>>>>> am I
>>>>>>>>>>>>> missing something?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Thanks,
>>>>>>>>>>>>> John Halley Gotway
>>>>>>>>>>>>> johnhg at ucar.edu
>>>>>>>>>>>>>
>>>>>>>>>>>>> Waylon.Collins wrote:
>>>>>>>>>>>>>  
>>>>>>>>>>>>>                                             
>>>>>>>>>>>>>                        
>>>>>>>>>>>>>> Dear Sir or Madam,
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> My goal is to generate verification statistics using the
>>>>>>>>>>>>>>                                                      
>>>>>>>>>>>>>>                           
>>>>>>>>>>> Point-Stat Tool
>>>>>>>>>>>                                                     
>>>>>>>>>>>>>> by comparing point observations with WRF-ARW gridded forecast
>>>>>>>>>>>>>> output.
>>>>>>>>>>>>>> With respect to the Point-Stat Tool, I understand that the
>>>>>>>>>>>>>> netCDF
>>>>>>>>>>>>>> obs_file should be created from pb2nc or ascii2nc. I do not
>>>>>>>>>>>>>> have
>>>>>>>>>>>>>> access
>>>>>>>>>>>>>> to data in prepbufr format, thus ascii2nc would be used.
>>>>>>>>>>>>>> However,                                                      
>>>>>>>>>>>>>>                           
>>>>>>>>>>> I will
>>>>>>>>>>>                                                     
>>>>>>>>>>>>>> be preparing verification statistics based on 3 months of
>>>>>>>>>>>>>> data,
>>>>>>>>>>>>>> every
>>>>>>>>>>>>>> 3-months, time resolution of 3 hours, for 5-10 observation
>>>>>>>>>>>>>> points.
>>>>>>>>>>>>>> Manually creating files to match the 10 column ASCII Point
>>>>>>>>>>>>>> Observation
>>>>>>>>>>>>>> Format as input into ascii2nc is not practical. I need an
>>>>>>>>>>>>>> efficient
>>>>>>>>>>>>>> method to create the obs_file as input into point_stat. I
>>>>>>>>>>>>>> have
>>>>>>>>>>>>>>                                                      
>>>>>>>>>>>>>>                           
>>>>>>>>>>> access to
>>>>>>>>>>>                                                     
>>>>>>>>>>>>>> the observations in netCDF format used by AWIPS. Attached is
>>>>>>>>>>>>>> the
>>>>>>>>>>>>>> output
>>>>>>>>>>>>>> from the linux command "ncdump -h" applied to one of these
>>>>>>>>>>>>>> files.                                                      
>>>>>>>>>>>>>>                           
>>>>>>>>>>> Do you
>>>>>>>>>>>                                                     
>>>>>>>>>>>>>> have software that will either (1) convert my AWIPS netCDF
>>>>>>>>>>>>>> files
>>>>>>>>>>>>>> into
>>>>>>>>>>>>>> the netCDF flavor required by point_stat, or (2) input my
>>>>>>>>>>>>>> AWIPS
>>>>>>>>>>>>>> netCDF
>>>>>>>>>>>>>> files and output an ASCII file containing the data in the 10
>>>>>>>>>>>>>> column
>>>>>>>>>>>>>> ASCII Point Observation Format required by ascii2nc?
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Waylon
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> ------------------------------------------------------------------------
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>>> Met_help mailing list
>>>>>>>>>>>>>> Met_help at mailman.ucar.edu
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>                                                          
>>>>>>>>>>>>>>                           
>>>>>>>>> _______________________________________________
>>>>>>>>> Met_help mailing list
>>>>>>>>> Met_help at mailman.ucar.edu
>>>>>>>>> http://mailman.ucar.edu/mailman/listinfo/met_help
>>>>>>>>>                                         
> 


More information about the Met_help mailing list