[Met_help] [rt.rap.ucar.edu #57975] History for building cluster in MODE

[John Halley Gotway via RT]

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

Hello,
i work with my chief on MetV3.0/MODE, and for now, we try to understand
how are build the clusters. I understand all about objects, how are used
the several threshold, the interest, the weight and function used to
build and match them. but once objects are defined and matched/merged, i
find nothing about cluster's building. Neither on MODE documentation,
nor in scientific papers. Only in the code (of course!), which is much
complicated and i would pass many month to understand only this part. Of
course, i know how to build clusters if i would have to do that. But the
question is not to write a code to calculate clusters, it is HOW IS IT
DONE IN MODE ? Could you give me some information, or a way to find it ?
Thank you very much,
--- Jean --- 

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

Subject: building cluster in MODE
From: John Halley Gotway
Time: Thu Aug 23 08:15:04 2012

Hello Jean,

Chris Davis forwarded your question about defining clusters in MODE to
me.  I work at NCAR on the MET development team.  We answer questions
about the use of MET, including MODE, via the MET-Help system:
    www.dtcenter.org/met/users/support/met_help.php

Please feel free to direct any future questions about the installation
and use of MET/MODE to met_help at ucar.edu.  I went ahead and created a
MET-Help ticket for the question you sent Chris.

It sounds like you understand the method using by MODE to define
objects.  We call the individual objects defined in the forecast and
observation fields "simple" objects.  Once you have simple objects in
each field, MODE applies additional logic (that's configurable via the
config file) to "merge" objects within the same field and "match"
objects between fields.

Merging may be done in up to 3 ways:
 (1) Applying the double-thresholding technique.  If two simple
objects created with the object definition threshold would be joined
together if a lower threshold was used, merge them together.
 (2) Applying fuzzy engine merging.  Use a fuzzy engine to compare all
the objects within one field to themselves.  If any matches are
created (other than each object matching itself), merge them together.
 (3) Applying the matching/merging logic.  When doing the matching
between fields, if two objects in one field match the same object in
the other field, merge those two objects together.

After one or all of those merging techniques have been applied, we end
up with groups of simple objects in one field that are associated with
groups of simple objects in the other field.  We calls those groups
"clusters".  A cluster is just a group of one or more simple objects
that matches a group of one or more simple objects in the other field.

Even through a cluster may contain multiple simple objects that are
disconnected, MODE treats it as a regular object.  The centroid, axis
angle, and all of the other object attributes are recomputed for the
cluster and are written out in the objects statistics line.

I realize that there are a lot of details here, and it can be pretty
confusing.  Please feel free to write with additional clarifying
questions.

With MODE, I often find it helpful to look at the PostScript output.
If you have specific questions about a particular run of MODE, please
send me the PostScript output and object statistics file, and I'd be
happy to try to explain the logic behind it.

Thanks,
John Halley Gotway
met_help at ucar.edu

------------------------------------------------
Subject: Re: [Fwd: [rt.rap.ucar.edu #57975] building cluster in MODE]
From: Jean.Barckicke at meteo.fr
Time: Fri Aug 24 03:50:00 2012

joel.stein a écrit :
>
>
> -------- Message original --------
> Sujet : 	[rt.rap.ucar.edu #57975] building cluster in MODE
> Date : 	Thu, 23 Aug 2012 08:15:04 -0600
> De : 	John Halley Gotway via RT <met_help at ucar.edu>
> Répondre à : 	met_help at ucar.edu
> Pour : 	Jean.Barckicke at meteo.fr
> Copie à : 	Joel.Stein at meteo.fr, cdavis at ucar.edu
> Références : 	<RT-Ticket-57975 at rap.ucar.edu>
> <rt-4.0.6-10713-1345729648-171.57975-6-0 at rap.ucar.edu>
>
>
>
> Hello Jean,
>
> Chris Davis forwarded your question about defining clusters in MODE
to me.  I work at NCAR on the MET development team.  We answer
questions about the use of MET, including MODE, via the MET-Help
system:
>     www.dtcenter.org/met/users/support/met_help.php
>
> Please feel free to direct any future questions about the
installation and use of MET/MODE to met_help at ucar.edu.  I went ahead
and created a MET-Help ticket for the question you sent Chris.
>
> It sounds like you understand the method using by MODE to define
objects.  We call the individual objects defined in the forecast and
observation fields "simple" objects.  Once you have simple objects in
each field, MODE applies additional logic (that's configurable via the
config file) to "merge" objects within the same field and "match"
objects between fields.
>
> Merging may be done in up to 3 ways:
>  (1) Applying the double-thresholding technique.  If two simple
objects created with the object definition threshold would be joined
together if a lower threshold was used, merge them together.
>  (2) Applying fuzzy engine merging.  Use a fuzzy engine to compare
all the objects within one field to themselves.  If any matches are
created (other than each object matching itself), merge them together.
>  (3) Applying the matching/merging logic.  When doing the matching
between fields, if two objects in one field match the same object in
the other field, merge those two objects together.
>
> After one or all of those merging techniques have been applied, we
end up with groups of simple objects in one field that are associated
with groups of simple objects in the other field.  We calls those
groups "clusters".  A cluster is just a group of one or more simple
objects that matches a group of one or more simple objects in the
other field.
>
> Even through a cluster may contain multiple simple objects that are
disconnected, MODE treats it as a regular object.  The centroid, axis
angle, and all of the other object attributes are recomputed for the
cluster and are written out in the objects statistics line.
>
> I realize that there are a lot of details here, and it can be pretty
confusing.  Please feel free to write with additional clarifying
questions.
>
> With MODE, I often find it helpful to look at the PostScript output.
If you have specific questions about a particular run of MODE, please
send me the PostScript output and object statistics file, and I'd be
happy to try to explain the logic behind it.
>
> Thanks,
> John Halley Gotway
> met_help at ucar.edu
>
>
>
hello John,
Thank you. All is perfectly clear. The only point is that i though
that
clustering was another layer above merging. But you answer that
clustering IS merging. it was not clear in documentation, or maybe
have
i not read carefully enough. moreover, i don't have found on run that
cluster were exactly the result of matching/merging (by reading output
file and viewing postscript output). Certainly have i done on error.
With you clear explanation, i will look once on my output and
certainly
agree with information you gave about clustering.
Best Regards,
 --- Jean ---
PS: i was too rapid when writing first mail. i don't have presented
me.
Please excuse me. I work as engineer in control/monitoring team at
Meteo-france/Toulouse.

------------------------------------------------
Subject: Re: [Fwd: [rt.rap.ucar.edu #57975] building cluster in MODE]
From: John Halley Gotway
Time: Fri Aug 24 08:05:20 2012

Jean,

I'll pass along your comments to the person who maintains the MET
user's guide.

There are many options in the MODE configuration file that define the
rules you want to apply in the matching/merging process.  I'd suggest
starting with the default settings, but you're welcome to
tweak the how you'd like.  But it is sometimes difficult to figure out
why an object was or was not matched/merged in a particular case.  If
that comes up, just send me the postscript and object stats
files and I'd be happy to take a look.

Thanks,
John

On 08/24/2012 03:50 AM, Jean.Barckicke at meteo.fr via RT wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=57975 >
>
> joel.stein a écrit :
>>
>>
>> -------- Message original --------
>> Sujet : 	[rt.rap.ucar.edu #57975] building cluster in MODE
>> Date : 	Thu, 23 Aug 2012 08:15:04 -0600
>> De : 	John Halley Gotway via RT <met_help at ucar.edu>
>> Répondre à : 	met_help at ucar.edu
>> Pour : 	Jean.Barckicke at meteo.fr
>> Copie à : 	Joel.Stein at meteo.fr, cdavis at ucar.edu
>> Références : 	<RT-Ticket-57975 at rap.ucar.edu>
>> <rt-4.0.6-10713-1345729648-171.57975-6-0 at rap.ucar.edu>
>>
>>
>>
>> Hello Jean,
>>
>> Chris Davis forwarded your question about defining clusters in MODE
to me.  I work at NCAR on the MET development team.  We answer
questions about the use of MET, including MODE, via the MET-Help
system:
>>      www.dtcenter.org/met/users/support/met_help.php
>>
>> Please feel free to direct any future questions about the
installation and use of MET/MODE to met_help at ucar.edu.  I went ahead
and created a MET-Help ticket for the question you sent Chris.
>>
>> It sounds like you understand the method using by MODE to define
objects.  We call the individual objects defined in the forecast and
observation fields "simple" objects.  Once you have simple objects in
each field, MODE applies additional logic (that's configurable via the
config file) to "merge" objects within the same field and "match"
objects between fields.
>>
>> Merging may be done in up to 3 ways:
>>   (1) Applying the double-thresholding technique.  If two simple
objects created with the object definition threshold would be joined
together if a lower threshold was used, merge them together.
>>   (2) Applying fuzzy engine merging.  Use a fuzzy engine to compare
all the objects within one field to themselves.  If any matches are
created (other than each object matching itself), merge them together.
>>   (3) Applying the matching/merging logic.  When doing the matching
between fields, if two objects in one field match the same object in
the other field, merge those two objects together.
>>
>> After one or all of those merging techniques have been applied, we
end up with groups of simple objects in one field that are associated
with groups of simple objects in the other field.  We calls those
groups "clusters".  A cluster is just a group of one or more simple
objects that matches a group of one or more simple objects in the
other field.
>>
>> Even through a cluster may contain multiple simple objects that are
disconnected, MODE treats it as a regular object.  The centroid, axis
angle, and all of the other object attributes are recomputed for the
cluster and are written out in the objects statistics line.
>>
>> I realize that there are a lot of details here, and it can be
pretty confusing.  Please feel free to write with additional
clarifying questions.
>>
>> With MODE, I often find it helpful to look at the PostScript
output.  If you have specific questions about a particular run of
MODE, please send me the PostScript output and object statistics file,
and I'd be happy to try to explain the logic behind it.
>>
>> Thanks,
>> John Halley Gotway
>> met_help at ucar.edu
>>
>>
>>
> hello John,
> Thank you. All is perfectly clear. The only point is that i though
that
> clustering was another layer above merging. But you answer that
> clustering IS merging. it was not clear in documentation, or maybe
have
> i not read carefully enough. moreover, i don't have found on run
that
> cluster were exactly the result of matching/merging (by reading
output
> file and viewing postscript output). Certainly have i done on error.
> With you clear explanation, i will look once on my output and
certainly
> agree with information you gave about clustering.
> Best Regards,
>   --- Jean ---
> PS: i was too rapid when writing first mail. i don't have presented
me.
> Please excuse me. I work as engineer in control/monitoring team at
> Meteo-france/Toulouse.
>


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