[Met_help] [rt.rap.ucar.edu #41465] History for question about "thresh"

[RAL HelpDesk {for John Halley Gotway}]

----------------------------------------------------------------
  Initial Request
----------------------------------------------------------------

Dear MET users,
 
I'm a new user of MET.
According to the Users' Guider, we can specify the threshold by using Fortran conventions.
For example, we can use "gt50" to specify the forcast field greate than 50. However,  I was
stumped by how to specify the  forcast field between 50 and 100. Because I want to evaluate
the model performance on modeling rainstorm (e.g. precipitation between 50 and 100mm).
 
best regards!
Feng Chen
 

----------------------------------------------------------------
  Complete Ticket History
----------------------------------------------------------------

Subject: Re: [rt.rap.ucar.edu #41465] question about "thresh"
From: John Halley Gotway
Time: Tue Oct 12 09:13:09 2010

Feng,

There's a few different ways to approach this issue.

First, when verifying precipitation, people usually choose a single
threshold of interest to define an "event".  For example, you may want
to define an "event" as being more than 50mm of
precipitation, i.e. "ge50.0".  You'd use that single threshold to
define a 2x2 contingency table, from which you can derive many
statistics (like probability of detection, critical success index, and
so on.  These are all contained in the "CTS" line that's output by
MET.

So, when you set the following in the Point-Stat or Grid-Stat config
files:
fcst_field[] = [ "ge50.0", "ge100.0" ];
The MET tools will use each of those thresholds listed to define a 2x2
contingency table, and derive a bunch of statistics for each.  The
output is stored in the CTC and CTS line types from the MET
tools.  See tables 4-4 and 4-5 of the MET User's Guide for the
contents of these lines.  This seems by far to be the most common way
of verifying precipitation.

Second, you mentioned that you're more interested in looking at a
range of values - for example, between 50mm and 100mm.  New in
METv3.0, we've added support for multi-dimensional contingency tables.
 The multi-dimensional contingency table output is stored in the MCTC
and MCTS line types.  See tables 4-7 and 4-8 of the MET User's Guide
for the contents of these lines.  As long as you have the
corresponding output_flag entries for the MCTC and MCTS line types
turned on, the MET tools will also use the thresholds you've provided
to define a multi-dimensional contingency table.  In our
example above (fcst_field[] = [ "ge50.0", "ge100.0" ];), it'll define
a 3x3 table.  All of the points with values less than 50 will go into
the first bin, between 50 and 100 will go into the second
bin, and greater than 100 will go into the third bin.  Unfortunately
though, the number of statistics you can derive from a 3x3 table is
much smaller than what you can derive from a 2x2 table.  In
table 4-8 of the MET User's Guide, you'll see that it's only accuracy,
hanssen-kuipers discriminant, heidke skill score, and gerrity score.

The last thing I want to tell you about is an option in defining the
verification "area" in the MET tools.  When you run Point-Stat and
Grid-Stat, you have to define the geographic region over which
you want to compute statistics.  Typically, this is either your whole
domain or some subset you've defined using a lat/lon polyline.
However, if you look at the comments in the config file for the
"mask_poly" parameter, you'll see there's another option that we call
"data masking".  You can pass a gridded data field to the tool, along
with a single threshold to be applied to that data field.
And only those grid points where the threshold criteria is met are
included in the verification.  One good example of this would be to
use a topography field, and only look at points greater than
2000m.  However, you're welcome to use the data masking however you'd
like.  For example, if you want to disregard any grid points where the
precipitation is less than 50mm, you could use the data
masking option to pass in that precipitation field as well the
threshold "ge50.0".

Hope that helps clarify your options in MET.

Thanks,
John Halley Gotway
met_help at ucar.edu

On 10/12/2010 03:18 AM, RAL HelpDesk {for fengfengabc004} wrote:
>
> Tue Oct 12 03:18:08 2010: Request 41465 was acted upon.
> Transaction: Ticket created by fengfengabc004 at 163.com
>        Queue: met_help
>      Subject: question about "thresh"
>        Owner: Nobody
>   Requestors: fengfengabc004 at 163.com
>       Status: new
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=41465 >
>
>
> Dear MET users,
>
> I'm a new user of MET.
> According to the Users' Guider, we can specify the threshold by
using Fortran conventions.
> For example, we can use "gt50" to specify the forcast field greate
than 50. However,  I was
> stumped by how to specify the  forcast field between 50 and 100.
Because I want to evaluate
> the model performance on modeling rainstorm (e.g. precipitation
between 50 and 100mm).
>
> best regards!
> Feng Chen
>

------------------------------------------------