[Met_help] [rt.rap.ucar.edu #57994] History for grid-stat and ensemble forecasts

[John Halley Gotway via RT]

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

Hi,
 
I am using MET4.0 to verify gridded forecasts of precipitation over the
CONUS. 
Verification and forecasts are in the same grid and in netcdf.
The forecasts  were initialized 4 times per day (00, 06, 12 and 18UTC)
during 15 December 2004 - 8 January 2005. 
I have no problems running Grid-Stat on the entire sample of forecasts, i.e.
forecasts are treated as deterministic.
 
However, I would like to consider the 4 initializations per day as an
ensemble (i.e. 4 members for each day). It is not clear how to do it since I
am only performing gridded verification. Should I use Ensemble Stat?
 
The other question is: can I still use Stat-Analysis tool to summarize the
outputs of Grid-Stat even though I am not running Point-Stat and Wavelet
tools?
 
Thanks again for the help,
 
 
Charles.
 
*====================================================*
Dr. Charles Jones
Researcher
Earth Research Institute (ERI)
University of California, Santa Barbara, CA 93106-3060
 
phone: (805) 893-5824
fax:   (805) 893-2578
email:  <mailto:cjones at icess.ucsb.edu> cjones at eri.ucsb.edu
URL:   <http://www.icess.ucsb.edu/asr/> http://www.eri.ucsb.edu/asr/
|=====================================================*
 


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

Subject: Re: [rt.rap.ucar.edu #57994] grid-stat and ensemble forecasts
From: John Halley Gotway
Time: Fri Aug 24 08:40:25 2012

Charles,

First of all, I'm glad you're able to get Grid-Stat up and running
well on your data.

Based on your description, it sounds like you'd like to create a time-
lagged ensemble forecast using data from 4 daily initializations.
Your guess is correct, you can use the Ensemble-Stat tool in
this case.  Unlike Point-Stat and Grid-Stat, we set up the Ensemble-
Stat tool to handle both gridded and point observations.  So you can
use it with your gridded observations.

But let me give you an overview of the Ensemble-Stat functionality.
Ensemble-Stat does two main things.  First, it reads in forecast
fields from the data files you pass it, and it derives basic
ensemble fields - like a simple ensemble mean, standard deviation, and
a relative frequency of meeting some threshold criteria you've defined
in the config file (these are like probability forecasts).
  Let me warn you though that Ensemble-Stat is not intended as a
comprehensive ensemble post-processing tool.  All it's giving you is
the most basic ensemble mean.  There are a lot of other more
sophisticated ways of generating ensemble products that MET doesn't
do.  The second part of Ensemble-Stat is comparing the input forecast
fields to the observations you've passed it.  The output of
this step is the computation of rank histograms - a useful way of
evaluating ensemble performance.

So here's how you'd use Ensemble-Stat.  Run it to generate an ensemble
mean field as well as rank histograms.  Then pass the NetCDF ensemble-
mean output file to Grid-Stat and verify it against those
same observations to generate statistics like you've done already.  In
this way, Ensemble-Stat is a "pre-processing" tool in that it
generates and ensemble mean and a "statistics" tool in that it
generates rank histograms.

As for the time-lagging part of it, you control that by what data
files you pass to the tool.  For each valid time (t) you run it,
you'll probably want to pass the 6-hour forecast from the t-6
initialization, the 12-hour forecast from t-12, the 18-hour forecast
from t-18, and the 24-hour forecast from t-24.

Hope that helps.

John Halley Gotway
met_help at ucar.edu

On 08/23/2012 09:38 PM, cjones via RT wrote:
>
> Thu Aug 23 21:38:08 2012: Request 57994 was acted upon.
> Transaction: Ticket created by cjones at icess.ucsb.edu
>         Queue: met_help
>       Subject: grid-stat and ensemble forecasts
>         Owner: Nobody
>    Requestors: cjones at icess.ucsb.edu
>        Status: new
>   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57994 >
>
>
> Hi,
>
> I am using MET4.0 to verify gridded forecasts of precipitation over
the
> CONUS.
> Verification and forecasts are in the same grid and in netcdf.
> The forecasts  were initialized 4 times per day (00, 06, 12 and
18UTC)
> during 15 December 2004 - 8 January 2005.
> I have no problems running Grid-Stat on the entire sample of
forecasts, i.e.
> forecasts are treated as deterministic.
>
> However, I would like to consider the 4 initializations per day as
an
> ensemble (i.e. 4 members for each day). It is not clear how to do it
since I
> am only performing gridded verification. Should I use Ensemble Stat?
>
> The other question is: can I still use Stat-Analysis tool to
summarize the
> outputs of Grid-Stat even though I am not running Point-Stat and
Wavelet
> tools?
>
> Thanks again for the help,
>
>
> Charles.
>
> *====================================================*
> Dr. Charles Jones
> Researcher
> Earth Research Institute (ERI)
> University of California, Santa Barbara, CA 93106-3060
>
> phone: (805) 893-5824
> fax:   (805) 893-2578
> email:  <mailto:cjones at icess.ucsb.edu> cjones at eri.ucsb.edu
> URL:   <http://www.icess.ucsb.edu/asr/> http://www.eri.ucsb.edu/asr/
> |=====================================================*
>
>


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