[Met_help] Questions about applying thresholds in MET

[John Henderson]

John,

Thanks for the clarification. A few questions are embedded below:

John Halley Gotway wrote:
> John,
>
> Hopefully I can clear some of this up.
>
> In the MET configuration files, thresholds should be specified using the "gt, ge, lt, le, eq, ne" FORTRAN convention for specifying inequalities:
> fcst_thresh[] = [ "le5.0 gt5.0" ];
> and
> fcst_wind_thresh[] = [ "gt5.0" ];
>
> Perhaps the source of confusion is that when we write the output thresholds, we do use the more conventional notation of ">, >=, <, <=, ==, !=".  Perhaps that was a poor choice and led to your
> confusion.  If the documentation is misleading, please let me know where, and we can clarify it.
>   
Unfortunately, my confusion is getting worse as I can't find anywhere in 
the documentation about using the usual ">", e.g., notation. The 
comments in the config file ARE clear about using the fortran 
convention. Perhaps it was simply the output in the VL1L2 file that made 
me want to get away from the fortran convention in the config file...
> As you are aware, in the STAT output data, there are some columns that contain threshold information: FCST_THRESH, OBS_THRESH, and COV_THRESH
> However, these columns are only used in those line types to which they apply.  For instance, since continuous statistics and SL1L2 partial sums are computed over the raw fields, with generally no
> thresholding applied, the CNT and SL1L2 output line types contain "NA" in those columns.  The exception to this rule is when the "fcst_wind_thresh" and "obs_wind_thresh" parameters are used.  In that
> case, those thresholds are used to populate the FCST_THRESH and OBS_THRESH columns in the VL1L2 output lines.
>
> The forecast and observation thresholds do apply when computing categorical statistics - using contingency tables.  You apply those thresholds to create 0/1 fields for the forecasts and observations
> and then put those counts into a contingency table and derive statistics.  Therefore the FCST_THRESH and OBS_THRESH columns contain threshold values in the FHO, CTC, and CTS line types.
>
> Perhaps the confusion is what exactly we mean by "thresholding".  I'm guessing you're thinking of thresholding as a process of filtering forecast and observation values and only keeping the ones that
> meet the threshold criteria.  That's not really what we mean.  Instead, we're keeping all the matched pairs (fcst, obs pairs) and applying the thresholds to those matched pairs to fill up a
> contingency table.
>
> I actually did uncover a bug based on this issue you sent.  When I specify multiple thresholds for "fcst_wind_thresh":
>    fcst_wind_thresh[] = [ "gt2.0 ge3.0" ];
> I'm only seeing output for the first threshold.  I'll fix this in the code, and it'll be included in the upcoming release of METv3.0.
>   
I believe the same results can be obtained by following the instructions 
in your 2nd email - that is, by repeating the 33,34 grib codes and 
simply specifying a different threshold.
> Thanks for the reminder about MADIS data.  I'll put that back on our list.
>
> Please let me know if you continue to have questions.
>
> Thanks,
> John
>   

John