[Met_help] MET questions

[John Halley Gotway]

Cory,

Good news.

Regarding your second question about PRMSL, it's an easy fix.  In the config file "PointStatConfig_PPmsl", just change
vx_grib_code[] = [ "PRMSL/P0"];
... to ...
vx_grib_code[] = [ "PRMSL/Z0"];

Here's the reasoning -
P0 tells Point-Stat to look for observations which have a corresponding pressure level value of 0 (i.e. obs_arr[2] = p_level = 0 in the netCDF observations).  Of course, none of your observations have
a pressure level of 0!
Z0 tells Point-Stat that we've verifying at a vertical level rather than a pressure level.  When Z is set to 0, 2, or 10 (i.e. surface), Point-Stat will use any observations with a message type of
APDSFC or SFCSHP.  Currently, MET really only verifies at pressure levels or at the surface.  I'd be interested to hear if there'd ever be a need to verify at a vertical level other than the surface.

So I was able to run your PMSL forecast: nz12kmN_d2_08012112_0006.00000_sslp.grb
And compare to the your PMSL observations: nz_metar_08012118_PPmsl.nc
Point-Stat found 121 pairs of observations to use.  However, only 63 of those pairs contained valid data for both forecast and observation, so that's while you'll see a count of 63 in the CNT output
file from Point-Stat.

Regarding your first question, about Spearman's Rank Correlation being negative (around -3.3) I'm going to refer you to Lacey Holland, a scientist in our group.  I don't know much about that
statistic, but hopefully Lacey can let me know if there's a problem in the way MET computes it.

Hope this helps,
John Halley Gotway


Cory Davis wrote:
> Hi there,
> I have a few questions and concerns about MET and in particular the
> point_stat tool.  These issues are related to whether I am using 2D
> model output fields correctly, negative returned values for the Spearman
> rank correlation coefficient, and trouble verifying mean sea level
> pressure.
> 
> I have managed to write python code which converts our METAR
> observations (already in netcdf) into MET compatible netcdf files.
> Also, our operational system already produces grib output via RIP for
> several variables, for instance 2m temperature, 10m windspeed, and mean
> sea level pressure, and these should correspond directly to the METAR
> observations.  I got the impression from the MET documentation that it
> is really expecting 3D model fields. However point_stat does run with
> these 2D model files, and also produces sensible output, at least for 2m
> temperature and 10m wind speed.  However, there does seem to be a
> problem with the Spearman rank correlation coefficient.  It is negative.
> Given that all the other returned values seem reasonable - the first
> impression is that this is a bug or another architecture dependent
> problem.  I have attached, obs., model, config, and output that should
> reproduce this.
> 
> For the mean sea level pressure I have a different problem: point_stat
> detects no pairs and hence no statistics.  I don't know why this should
> be the case given that the obs. and model files were produced by the
> same process as the T2 and windspeed files.  It is possibly an issue
> about how one specifies the level in the PointStatConfig_PPmsl file.
> Also, I will have a closer look at the grib file to see if this is
> somehow different to the T2 and ws10 grib files.
> 
> Any help with this would be hugely appreciated.
> 
> Cheers,
> Cory.
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Met_help mailing list
> Met_help at mailman.ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/met_help