[Met_help] Re: MET error

[John Halley Gotway]

Luke,

The accumulation interval is meant for precipitation fields.  If you use a utility like 'wgrib' to dump the contents of a grib file you may find precipitation fields with different accumulation
intervals.  For example, a single grib file may contain accumulated precipitation (APCP or grib code 61) with accumulation intervals of 1 hour and 24 hours.  So specifying "61/A24" or equivalently
"APCP/A24" would select out the 24 hour accumulated precip field.

I suppose you could try to use point_stat to compare precip to gauge data, but I'm really not sure if gauge data is available in the PrepBufr files.  If you'd like to compare gridded precip to gridded
observations like StageII or StageIV analysis, you could use the grid_stat or mode tools.

As for the observation files, in the MET beta release we're only supporting point observation data in PrepBufr format.  We realize that this is a limitation, and we've received requests from users to
support additional point observation formats.  We'll be adding support for the ObsProc (formatted ascii) observation format used by data assimilation folks in the near future.  And we may add support
for additional observation formats based on user feedback.

I do know that the PrepBufr data available from NCEP contains observations from outside of CONUS.  But as for the appropriateness of using the NCDC data, I'll defer to a scientist in our group, Lacey
Holland, who I've copied on this email.

Thanks,
John

Luke Peffers wrote:
> Hello John, I have a couple quick questions for you regarding my ongoing MET
> experimentation:
> 
> In the MET point_stat config file, there is an option for comparing
> observations of accumulation intervals (i.e. ANNN).  I am unclear as to what
> the accumulation interval should be set to for comparing various
> observations.  What could this be used for and how does one determine what
> to set the interval, ANNN, to?
> 
> Also, I am interested in WRF verification in non CONUS regions of the
> world.  I have found a PREPBUFR dataset from NCDC (DSI-6116
> http://www1.ncdc.noaa.gov/pub/data/documentlibrary/tddoc/td6116.pdf ) ,
> which I hope will include enough observations to make a sound statistical
> comparison with my WRF run. I would like to know if you are aware of any
> other global datasets that can be used with MET point-stat and if not, what
> do you think of a dataset such as the above mentioned?
> 
> Thank you for all your time and assistance.
> 
> Luke Peffers
> FSU
> 
> On 10/5/07, Luke Peffers <luke.peffers at gmail.com> wrote:
>> Thanks John, I realized that there may be an issue with sample size since
>> I am trying to verify at 6Z.  I now know to use the most robust  sounding
>> data which is included in the 00Z and 12Z obs.  My mistake, but I'm glad my
>> amateurism is contributing to your work!  I'm happy to help.
>>
>> Luke
>>
>>
>> On 10/5/07, John Halley Gotway <johnhg at rap.ucar.edu> wrote:
>>> Luke,
>>>
>>> This is good stuff.  I'm glad you caught this error.  What's happening
>>> is that we're computing statistics for a sample size of only 1!  When we go
>>> to compute confidence intervals, needless to say,
>>> there are problems.  I've added checks to only compute confidence
>>> intervals when the sample size is large enough to do so.
>>>
>>> However, we're leaving it up to the users to interpret their
>>> results.  For example, if you have a confidence interval computed for a very
>>> small sample size (under 10), I wouldn't trust it very much.
>>>
>>> I've attached an updated version of the file "point_stat.cc".  You can
>>> use it to replace the version in METv0.9/src/point_stat.  Then at the
>>> top-level, do a "make clean" followed by a "make >&
>>> make_met.log".
>>>
>>> Let me know if you run into other problems.  I'll include these changes
>>> in the development version of MET so that they appear in the next official
>>> release.
>>>
>>> Thanks for finding this bug Luke.  We really appreciate it.
>>>
>>> John
>>>
>>> Luke Peffers wrote:
>>>> Ok John, I have "put" the data you requested in your directory via
>>> ftp.  As
>>>> for the point_stat statement, I did enter it correctly but goofed when
>>> I
>>>> typed the email.  Too bad that wasn't the only problem.  Please let me
>>> know
>>>> what you find.
>>>>
>>>> Luke Peffers
>>>> FSU
>>>>
>>>> On 10/4/07, John Halley Gotway <johnhg at rap.ucar.edu> wrote:
>>>>> Luke,
>>>>>
>>>>> I haven't seen that error message before.  One thing to point out
>>>>> though... If you look at the usage statement for point_stat, you'll
>>> see
>>>>> that it's expecting the "model" name to be listed as the 3rd
>>> argument.  So
>>>>> you should running the command as:
>>>>> /point_stat ./WRFPRS_d01.72  ./OBS_10_01_06.nc WRF_SingleG -config
>>>>> PointStatConfig_default
>>>>>
>>>>> Otherwise, it'll think the model name is "-config".  I should
>>> probably
>>>>> make the argument processing smarter than that to skip the optional
>>>>> arguments.
>>>>>
>>>>> Anyway, I'm guessing I'll need to run the data you're using to try to
>>>>> reproduce the error.  Would you be able to post those files to the
>>> FTP
>>>>> site again?
>>>>>
>>>>> I'd need the 3 files:
>>>>> WRFPRS_d01.72
>>>>> OBS_10_01_06.nc
>>>>> PointStatConfig_default
>>>>>
>>>>> Please let me know when they're up there, and I'll give it a shot.
>>>>>
>>>>> Thanks,
>>>>> John
>>>>>
>>>>>> Hello John:
>>>>>>
>>>>>> I have another error for you.  I received this one with this
>>> command:
>>>>>> ./point_stat ./WRFPRS_d01.72     ./OBS_10_01_06.nc     -config
>>>>>> ./PointStatConfig_default     WRF_SingleG
>>>>>>
>>>>>> gsl: beta.c:56: ERROR: domain error
>>>>>> Default GSL error handler invoked.
>>>>>> Abort
>>>>>>
>>>>>> Do you know where things may have gone wrong?
>>>>>>
>>>>>> Luke Peffers
>>>>>> FSU
>>>>>>
>>>>>
>>>
>