[Met_help] [rt.rap.ucar.edu #86937] History for MTD object question
John Halley Gotway via RT
met_help at ucar.edu
Wed Sep 12 16:19:05 MDT 2018
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
I am running into a difficulty with defining objects in MTD.
The issue comes up when at one time step there are 2 2-d objects that are not that close to each other, and then at the next time step they are close. MTD is calling that 1 object, and I would like for it to be 2.
I have conv_radius set to 0. Is there something else I can set to encourage MTD to define 2 objects in that situation?
Thanks!
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: MTD object question
From: John Halley Gotway
Time: Wed Sep 12 12:43:02 2018
Hi Laurel,
I moved your question over to the met-help ticketing system and am
responding through that. While you ask a pretty simple, straight-
forward question, it's a bit difficult to answer. Those of us that
have run traditional MODE in the past are used to thinking about 2-
dimensional. But MTD adds the time dimension and that requires a bit
of a paradigm shift.
I read that your data has objects that are distinct and well-separated
at one time and then distinct but closer at the next time. They are
currently being included in the same MTD space/time object but you
would prefer that they aren't. Here's two details that may help
explain what's happening:
(1) If two objects are distinct at one time but eventually overlap at
a future time, then they will, by definition, be included in the same
MTD space/time objects. MTD handles these cases of "merging" (2 or
more objects become 1)... as well as "splitting" (1 object becomes 2
or more).
Take a close look at your output to see if those 2 objects eventually
meet. If so, then there's no way of keeping them apart.
(2) Traditional MODE supports matching/merging logic which states that
if two object in one field match the same object in the other, then
those two objects are merged. That same logic is inherited in MTD.
If two MTD space-time objects match in one field match the same MTD
space-time object in the other field, then they're merged.
Take a close look at your output to see if that's occurring.
In the MTD config file, you can adjust the fuzzy logic attribute
weights and threshold... but that only impacts the logic described in
(2).
If you have additional questions, we can ask Randy Bullock, the
developer of MTD, to clarify things. In that case, it'd be helpful to
have you send us some sample MTD output to illustrate the question.
Thanks,
John Halley Gotway
On Wed Sep 12 11:28:05 2018, johnhg wrote:
> I am running into a difficulty with defining objects in MTD.
> The issue comes up when at one time step there are 2 2-d objects
that
> are not that close to each other, and then at the next time step
they
> are close. MTD is calling that 1 object, and I would like for it to
> be 2.
> I have conv_radius set to 0. Is there something else I can set to
> encourage MTD to define 2 objects in that situation?
>
> Thanks!
------------------------------------------------
Subject: MTD object question
From: Dehaan, Laurel
Time: Wed Sep 12 13:17:36 2018
Thank you for your prompt reply John!
I will check closer if the 2 objects do overlap. It occurred to me
(reading your email) that it's quite likely that they do overlap if I
look
at a 3rd time step.
I had a couple follow up questions.
-Is it true that if 2 objects never overlap, that setting conv radius
to 0
should prevent them from merging? (ie - since MTD is defining a
single
object and conv radius = 0, must the objects overlap at some point?)
-Your second point gets to the root of the problem. What is happening
is
that there are 2 atmospheric rivers, one right after the other. In
the
forecast the 2nd atmospheric river starts a time step earlier than the
observation - which causes MTD to make the forecasted AR one large
object
and the observed AR 2 smaller objects. All of these objects are
unmatched
- I assume because the length of time is so different.
So, reducing the time weight in the fuzzy logic or increasing the
"tails"
in time centroid delta would increase the chance that the objects
would
match, true? Is there anything else I could do to match the objects
(either as 1 or 2) - since I had other problems with a time weight
that was
too low.
Thanks again for your help!
-Laurel
On Wed, Sep 12, 2018 at 11:43 AM John Halley Gotway via RT <
met_help at ucar.edu> wrote:
> Hi Laurel,
>
> I moved your question over to the met-help ticketing system and am
> responding through that. While you ask a pretty simple, straight-
forward
> question, it's a bit difficult to answer. Those of us that have run
> traditional MODE in the past are used to thinking about 2-
dimensional. But
> MTD adds the time dimension and that requires a bit of a paradigm
shift.
>
> I read that your data has objects that are distinct and well-
separated at
> one time and then distinct but closer at the next time. They are
currently
> being included in the same MTD space/time object but you would
prefer that
> they aren't. Here's two details that may help explain what's
happening:
>
> (1) If two objects are distinct at one time but eventually overlap
at a
> future time, then they will, by definition, be included in the same
MTD
> space/time objects. MTD handles these cases of "merging" (2 or more
> objects become 1)... as well as "splitting" (1 object becomes 2 or
more).
>
> Take a close look at your output to see if those 2 objects
eventually
> meet. If so, then there's no way of keeping them apart.
>
> (2) Traditional MODE supports matching/merging logic which states
that if
> two object in one field match the same object in the other, then
those two
> objects are merged. That same logic is inherited in MTD. If two
MTD
> space-time objects match in one field match the same MTD space-time
object
> in the other field, then they're merged.
>
> Take a close look at your output to see if that's occurring.
>
> In the MTD config file, you can adjust the fuzzy logic attribute
weights
> and threshold... but that only impacts the logic described in (2).
>
> If you have additional questions, we can ask Randy Bullock, the
developer
> of MTD, to clarify things. In that case, it'd be helpful to have
you send
> us some sample MTD output to illustrate the question.
>
> Thanks,
> John Halley Gotway
>
>
>
>
> On Wed Sep 12 11:28:05 2018, johnhg wrote:
> > I am running into a difficulty with defining objects in MTD.
> > The issue comes up when at one time step there are 2 2-d objects
that
> > are not that close to each other, and then at the next time step
they
> > are close. MTD is calling that 1 object, and I would like for it
to
> > be 2.
> > I have conv_radius set to 0. Is there something else I can set to
> > encourage MTD to define 2 objects in that situation?
> >
> > Thanks!
>
>
>
>
------------------------------------------------
Subject: MTD object question
From: John Halley Gotway
Time: Wed Sep 12 14:15:17 2018
Laurel,
*-Is it true that if 2 objects never overlap, that setting conv radius
to 0*
* should prevent them from merging? (ie - since MTD is defining a
single*
* object and conv radius = 0, must the objects overlap at some
point?)*
Generally yes, it's true that if two 2D time slices never overlap then
they
should be in separate 3D space-time objects. Setting the conv_radius
to 0
effectively disables the spatial smoothing. BUT, and this is a big
caveat,
MTD also includes a hard-coded temporal smoother. The convolved value
for
a single grid point is computed by averaging values over the spatial
area
(which is just that one point when conv_radius = 0) **AND** the values
for
one time step before/after the current time. If you can imagine one
object
moving across the grid with a second object following it, always one
grid
square behind... I think those would get merged because of the
temporal
smoothing.
And there currently is no way in MTD to configure how many timestep
+/-
should be used for that.
Yes, you can modify the fuzzy interest weights and/or total interest
threshold to play with whether objects match between the forecast and
observation.
John
On Wed, Sep 12, 2018 at 1:18 PM Dehaan, Laurel via RT
<met_help at ucar.edu>
wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=86937 >
>
> Thank you for your prompt reply John!
>
> I will check closer if the 2 objects do overlap. It occurred to me
> (reading your email) that it's quite likely that they do overlap if
I look
> at a 3rd time step.
>
> I had a couple follow up questions.
> -Is it true that if 2 objects never overlap, that setting conv
radius to 0
> should prevent them from merging? (ie - since MTD is defining a
single
> object and conv radius = 0, must the objects overlap at some point?)
>
> -Your second point gets to the root of the problem. What is
happening is
> that there are 2 atmospheric rivers, one right after the other. In
the
> forecast the 2nd atmospheric river starts a time step earlier than
the
> observation - which causes MTD to make the forecasted AR one large
object
> and the observed AR 2 smaller objects. All of these objects are
unmatched
> - I assume because the length of time is so different.
> So, reducing the time weight in the fuzzy logic or increasing the
"tails"
> in time centroid delta would increase the chance that the objects
would
> match, true? Is there anything else I could do to match the objects
> (either as 1 or 2) - since I had other problems with a time weight
that was
> too low.
>
> Thanks again for your help!
>
> -Laurel
>
> On Wed, Sep 12, 2018 at 11:43 AM John Halley Gotway via RT <
> met_help at ucar.edu> wrote:
>
> > Hi Laurel,
> >
> > I moved your question over to the met-help ticketing system and am
> > responding through that. While you ask a pretty simple, straight-
forward
> > question, it's a bit difficult to answer. Those of us that have
run
> > traditional MODE in the past are used to thinking about 2-
dimensional.
> But
> > MTD adds the time dimension and that requires a bit of a paradigm
shift.
> >
> > I read that your data has objects that are distinct and well-
separated at
> > one time and then distinct but closer at the next time. They are
> currently
> > being included in the same MTD space/time object but you would
prefer
> that
> > they aren't. Here's two details that may help explain what's
happening:
> >
> > (1) If two objects are distinct at one time but eventually overlap
at a
> > future time, then they will, by definition, be included in the
same MTD
> > space/time objects. MTD handles these cases of "merging" (2 or
more
> > objects become 1)... as well as "splitting" (1 object becomes 2 or
more).
> >
> > Take a close look at your output to see if those 2 objects
eventually
> > meet. If so, then there's no way of keeping them apart.
> >
> > (2) Traditional MODE supports matching/merging logic which states
that if
> > two object in one field match the same object in the other, then
those
> two
> > objects are merged. That same logic is inherited in MTD. If two
MTD
> > space-time objects match in one field match the same MTD space-
time
> object
> > in the other field, then they're merged.
> >
> > Take a close look at your output to see if that's occurring.
> >
> > In the MTD config file, you can adjust the fuzzy logic attribute
weights
> > and threshold... but that only impacts the logic described in (2).
> >
> > If you have additional questions, we can ask Randy Bullock, the
developer
> > of MTD, to clarify things. In that case, it'd be helpful to have
you
> send
> > us some sample MTD output to illustrate the question.
> >
> > Thanks,
> > John Halley Gotway
> >
> >
> >
> >
> > On Wed Sep 12 11:28:05 2018, johnhg wrote:
> > > I am running into a difficulty with defining objects in MTD.
> > > The issue comes up when at one time step there are 2 2-d objects
that
> > > are not that close to each other, and then at the next time step
they
> > > are close. MTD is calling that 1 object, and I would like for
it to
> > > be 2.
> > > I have conv_radius set to 0. Is there something else I can set
to
> > > encourage MTD to define 2 objects in that situation?
> > >
> > > Thanks!
> >
> >
> >
> >
>
>
------------------------------------------------
Subject: MTD object question
From: Dehaan, Laurel
Time: Wed Sep 12 14:29:09 2018
Thanks John.
That's very helpful
-Laurel
On Wed, Sep 12, 2018 at 1:15 PM John Halley Gotway via RT
<met_help at ucar.edu>
wrote:
> Laurel,
>
> *-Is it true that if 2 objects never overlap, that setting conv
radius to
> 0*
>
> * should prevent them from merging? (ie - since MTD is defining a
single*
> * object and conv radius = 0, must the objects overlap at some
point?)*
>
> Generally yes, it's true that if two 2D time slices never overlap
then they
> should be in separate 3D space-time objects. Setting the
conv_radius to 0
> effectively disables the spatial smoothing. BUT, and this is a big
caveat,
> MTD also includes a hard-coded temporal smoother. The convolved
value for
> a single grid point is computed by averaging values over the spatial
area
> (which is just that one point when conv_radius = 0) **AND** the
values for
> one time step before/after the current time. If you can imagine one
object
> moving across the grid with a second object following it, always one
grid
> square behind... I think those would get merged because of the
temporal
> smoothing.
>
> And there currently is no way in MTD to configure how many timestep
+/-
> should be used for that.
>
> Yes, you can modify the fuzzy interest weights and/or total interest
> threshold to play with whether objects match between the forecast
and
> observation.
>
> John
>
> On Wed, Sep 12, 2018 at 1:18 PM Dehaan, Laurel via RT
<met_help at ucar.edu>
> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=86937 >
> >
> > Thank you for your prompt reply John!
> >
> > I will check closer if the 2 objects do overlap. It occurred to
me
> > (reading your email) that it's quite likely that they do overlap
if I
> look
> > at a 3rd time step.
> >
> > I had a couple follow up questions.
> > -Is it true that if 2 objects never overlap, that setting conv
radius to
> 0
> > should prevent them from merging? (ie - since MTD is defining a
single
> > object and conv radius = 0, must the objects overlap at some
point?)
> >
> > -Your second point gets to the root of the problem. What is
happening is
> > that there are 2 atmospheric rivers, one right after the other.
In the
> > forecast the 2nd atmospheric river starts a time step earlier than
the
> > observation - which causes MTD to make the forecasted AR one large
object
> > and the observed AR 2 smaller objects. All of these objects are
> unmatched
> > - I assume because the length of time is so different.
> > So, reducing the time weight in the fuzzy logic or increasing the
"tails"
> > in time centroid delta would increase the chance that the objects
would
> > match, true? Is there anything else I could do to match the
objects
> > (either as 1 or 2) - since I had other problems with a time weight
that
> was
> > too low.
> >
> > Thanks again for your help!
> >
> > -Laurel
> >
> > On Wed, Sep 12, 2018 at 11:43 AM John Halley Gotway via RT <
> > met_help at ucar.edu> wrote:
> >
> > > Hi Laurel,
> > >
> > > I moved your question over to the met-help ticketing system and
am
> > > responding through that. While you ask a pretty simple,
> straight-forward
> > > question, it's a bit difficult to answer. Those of us that have
run
> > > traditional MODE in the past are used to thinking about 2-
dimensional.
> > But
> > > MTD adds the time dimension and that requires a bit of a
paradigm
> shift.
> > >
> > > I read that your data has objects that are distinct and well-
separated
> at
> > > one time and then distinct but closer at the next time. They
are
> > currently
> > > being included in the same MTD space/time object but you would
prefer
> > that
> > > they aren't. Here's two details that may help explain what's
> happening:
> > >
> > > (1) If two objects are distinct at one time but eventually
overlap at a
> > > future time, then they will, by definition, be included in the
same MTD
> > > space/time objects. MTD handles these cases of "merging" (2 or
more
> > > objects become 1)... as well as "splitting" (1 object becomes 2
or
> more).
> > >
> > > Take a close look at your output to see if those 2 objects
eventually
> > > meet. If so, then there's no way of keeping them apart.
> > >
> > > (2) Traditional MODE supports matching/merging logic which
states that
> if
> > > two object in one field match the same object in the other, then
those
> > two
> > > objects are merged. That same logic is inherited in MTD. If
two MTD
> > > space-time objects match in one field match the same MTD space-
time
> > object
> > > in the other field, then they're merged.
> > >
> > > Take a close look at your output to see if that's occurring.
> > >
> > > In the MTD config file, you can adjust the fuzzy logic attribute
> weights
> > > and threshold... but that only impacts the logic described in
(2).
> > >
> > > If you have additional questions, we can ask Randy Bullock, the
> developer
> > > of MTD, to clarify things. In that case, it'd be helpful to
have you
> > send
> > > us some sample MTD output to illustrate the question.
> > >
> > > Thanks,
> > > John Halley Gotway
> > >
> > >
> > >
> > >
> > > On Wed Sep 12 11:28:05 2018, johnhg wrote:
> > > > I am running into a difficulty with defining objects in MTD.
> > > > The issue comes up when at one time step there are 2 2-d
objects that
> > > > are not that close to each other, and then at the next time
step they
> > > > are close. MTD is calling that 1 object, and I would like for
it to
> > > > be 2.
> > > > I have conv_radius set to 0. Is there something else I can
set to
> > > > encourage MTD to define 2 objects in that situation?
> > > >
> > > > Thanks!
> > >
> > >
> > >
> > >
> >
> >
>
>
------------------------------------------------
More information about the Met_help
mailing list