[Met_help] [rt.rap.ucar.edu #57544] History for How to select convolution radius / threshold values in MODE

John Halley Gotway via RT met_help at ucar.edu
Mon Jul 23 10:54:33 MDT 2012


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

Hello,

Is there a standard way of selecting convolution radius/threshold values in
in the MODE" tool?
For example, for forecast/observation data with 40km grid spacing, what
are the typical radius/threshold values?

Many thanks in advance,

Endalk

-- 
-----------------------------------------------------------------------------------
Endalkachew Bekele
NOAA Climate Prediction Center / African Desk
5200 Auth Road, Rm 811, Camp Springs, MD 20746, USA
Tel: (301) 763 8000 Ext. 7014
e-mail: endalkachew.bekele at noaa.gov


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

Subject: Re: [rt.rap.ucar.edu #57544] How to select convolution radius / threshold values in MODE
From: John Halley Gotway
Time: Mon Jul 23 10:31:45 2012

Endalk,

Good question.  The answer is basically no, there is no "standard way"
of selecting the MODE convolution radius and threshold, but I can give
you some tips.

The process of selecting them is typically an iterative one.  Pick out
a case of interest, run it through MODE, and look at the PostScript
output.  If you're dealing with precipitation, I'd suggest
starting with a convolution radius of 5 and a threshold of ">0.0".
That will probably give you very large objects.

Next, ask yourself, what type of event are you really trying to
characterize with the MODE objects?  Are you interested in the
occurrence of any rain (>0.0 mm), moderate rain, or intense rain?
Choose
your threshold based on the event of interest, keeping in mind the
accumulation interval you're evaluating.  Try running again and look
at the PostScript output.  Do the objects defined capture the
events in the raw field you were trying to capture?

If not, you could try increasing the convolution radius to get a
smaller number of smoother/larger objects.  Or you could decrease it
to get a larger number of sharper/smaller objects.  You could turn
off the convolution step altogether by setting the radius to 0, but
you'll find that you have way too many small objects and MODE takes
too long to run keeping track of all that object information.

Just continue tweaking the convolution radius/threshold to define the
objects that match the phenomena you're trying to capture.  After each
run, look at the PostScript output to decided what to try next.

Sorry I don't have a more simple answer for you, but hopefully this
will get you going in the right direction.

One last note - I'd suggests choosing your parameters based on a few
cases of interest.  But once you have them set, you should keep them
fixed when running across a large dataset.  MODE objects that
are defined using different criteria really are not comparable.  So
you have to fix the MODE settings before doing an extended analysis.

Thanks,
John Halley Gotway
met_help at ucar.edu

On 07/23/2012 10:18 AM, Endalkachew Bekele via RT wrote:
>
> Mon Jul 23 10:18:41 2012: Request 57544 was acted upon.
> Transaction: Ticket created by endalkachew.bekele at noaa.gov
>         Queue: met_help
>       Subject: How to select convolution radius / threshold values
in MODE
>         Owner: Nobody
>    Requestors: endalkachew.bekele at noaa.gov
>        Status: new
>   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57544 >
>
>
> Hello,
>
> Is there a standard way of selecting convolution radius/threshold
values in
> in the MODE" tool?
> For example, for forecast/observation data with 40km grid spacing,
what
> are the typical radius/threshold values?
>
> Many thanks in advance,
>
> Endalk
>


------------------------------------------------
Subject: How to select convolution radius / threshold values in MODE
From: Endalkachew Bekele
Time: Mon Jul 23 10:48:37 2012

John,

Thank you so much for the tips.

I will go through the steps you suggested to find meaningful values.

Thanks again,

Endalk
On Mon, Jul 23, 2012 at 12:31 PM, John Halley Gotway via RT <
met_help at ucar.edu> wrote:

> Endalk,
>
> Good question.  The answer is basically no, there is no "standard
way" of
> selecting the MODE convolution radius and threshold, but I can give
you
> some tips.
>
> The process of selecting them is typically an iterative one.  Pick
out a
> case of interest, run it through MODE, and look at the PostScript
output.
>  If you're dealing with precipitation, I'd suggest
> starting with a convolution radius of 5 and a threshold of ">0.0".
That
> will probably give you very large objects.
>
> Next, ask yourself, what type of event are you really trying to
> characterize with the MODE objects?  Are you interested in the
occurrence
> of any rain (>0.0 mm), moderate rain, or intense rain?  Choose
> your threshold based on the event of interest, keeping in mind the
> accumulation interval you're evaluating.  Try running again and look
at the
> PostScript output.  Do the objects defined capture the
> events in the raw field you were trying to capture?
>
> If not, you could try increasing the convolution radius to get a
smaller
> number of smoother/larger objects.  Or you could decrease it to get
a
> larger number of sharper/smaller objects.  You could turn
> off the convolution step altogether by setting the radius to 0, but
you'll
> find that you have way too many small objects and MODE takes too
long to
> run keeping track of all that object information.
>
> Just continue tweaking the convolution radius/threshold to define
the
> objects that match the phenomena you're trying to capture.  After
each run,
> look at the PostScript output to decided what to try next.
>
> Sorry I don't have a more simple answer for you, but hopefully this
will
> get you going in the right direction.
>
> One last note - I'd suggests choosing your parameters based on a few
cases
> of interest.  But once you have them set, you should keep them fixed
when
> running across a large dataset.  MODE objects that
> are defined using different criteria really are not comparable.  So
you
> have to fix the MODE settings before doing an extended analysis.
>
> Thanks,
> John Halley Gotway
> met_help at ucar.edu
>
> On 07/23/2012 10:18 AM, Endalkachew Bekele via RT wrote:
> >
> > Mon Jul 23 10:18:41 2012: Request 57544 was acted upon.
> > Transaction: Ticket created by endalkachew.bekele at noaa.gov
> >         Queue: met_help
> >       Subject: How to select convolution radius / threshold values
in
> MODE
> >         Owner: Nobody
> >    Requestors: endalkachew.bekele at noaa.gov
> >        Status: new
> >   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57544 >
>  >
> >
> > Hello,
> >
> > Is there a standard way of selecting convolution radius/threshold
values
> in
> > in the MODE" tool?
> > For example, for forecast/observation data with 40km grid spacing,
what
> > are the typical radius/threshold values?
> >
> > Many thanks in advance,
> >
> > Endalk
> >
>
>
>


--
-----------------------------------------------------------------------------------
Endalkachew Bekele
NOAA Climate Prediction Center / African Desk
5200 Auth Road, Rm 811, Camp Springs, MD 20746, USA
Tel: (301) 763 8000 Ext. 7014
e-mail: endalkachew.bekele at noaa.gov

------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #57544] How to select convolution radius / threshold values in MODE
From: John Halley Gotway
Time: Mon Jul 23 10:54:16 2012

Endalk,

Great.  Just let us know if future questions come up in your use of
MET and MODE.

John

On 07/23/2012 10:48 AM, Endalkachew Bekele via RT wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57544 >
>
> John,
>
> Thank you so much for the tips.
>
> I will go through the steps you suggested to find meaningful values.
>
> Thanks again,
>
> Endalk
> On Mon, Jul 23, 2012 at 12:31 PM, John Halley Gotway via RT <
> met_help at ucar.edu> wrote:
>
>> Endalk,
>>
>> Good question.  The answer is basically no, there is no "standard
way" of
>> selecting the MODE convolution radius and threshold, but I can give
you
>> some tips.
>>
>> The process of selecting them is typically an iterative one.  Pick
out a
>> case of interest, run it through MODE, and look at the PostScript
output.
>>   If you're dealing with precipitation, I'd suggest
>> starting with a convolution radius of 5 and a threshold of ">0.0".
That
>> will probably give you very large objects.
>>
>> Next, ask yourself, what type of event are you really trying to
>> characterize with the MODE objects?  Are you interested in the
occurrence
>> of any rain (>0.0 mm), moderate rain, or intense rain?  Choose
>> your threshold based on the event of interest, keeping in mind the
>> accumulation interval you're evaluating.  Try running again and
look at the
>> PostScript output.  Do the objects defined capture the
>> events in the raw field you were trying to capture?
>>
>> If not, you could try increasing the convolution radius to get a
smaller
>> number of smoother/larger objects.  Or you could decrease it to get
a
>> larger number of sharper/smaller objects.  You could turn
>> off the convolution step altogether by setting the radius to 0, but
you'll
>> find that you have way too many small objects and MODE takes too
long to
>> run keeping track of all that object information.
>>
>> Just continue tweaking the convolution radius/threshold to define
the
>> objects that match the phenomena you're trying to capture.  After
each run,
>> look at the PostScript output to decided what to try next.
>>
>> Sorry I don't have a more simple answer for you, but hopefully this
will
>> get you going in the right direction.
>>
>> One last note - I'd suggests choosing your parameters based on a
few cases
>> of interest.  But once you have them set, you should keep them
fixed when
>> running across a large dataset.  MODE objects that
>> are defined using different criteria really are not comparable.  So
you
>> have to fix the MODE settings before doing an extended analysis.
>>
>> Thanks,
>> John Halley Gotway
>> met_help at ucar.edu
>>
>> On 07/23/2012 10:18 AM, Endalkachew Bekele via RT wrote:
>>>
>>> Mon Jul 23 10:18:41 2012: Request 57544 was acted upon.
>>> Transaction: Ticket created by endalkachew.bekele at noaa.gov
>>>          Queue: met_help
>>>        Subject: How to select convolution radius / threshold
values in
>> MODE
>>>          Owner: Nobody
>>>     Requestors: endalkachew.bekele at noaa.gov
>>>         Status: new
>>>    Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57544 >
>>   >
>>>
>>> Hello,
>>>
>>> Is there a standard way of selecting convolution radius/threshold
values
>> in
>>> in the MODE" tool?
>>> For example, for forecast/observation data with 40km grid spacing,
what
>>> are the typical radius/threshold values?
>>>
>>> Many thanks in advance,
>>>
>>> Endalk
>>>
>>
>>
>>
>
>


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


More information about the Met_help mailing list