[Met_help] [rt.rap.ucar.edu #46651] History for Wind direction in point_stat
RAL HelpDesk {for John Halley Gotway}
met_help at ucar.edu
Fri Jun 17 14:14:00 MDT 2011
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
Hi again,
Let me preface this by saying I realize we've discussed this before with
previous versions of MET, but it appears that the new version of
point_stat (v3.0) is much closer to being able to handle wind direction
verification. With that in mind, I wanted to ask a few questions and
run a few things by you to make sure I'm not barking up the wrong tree,
or over-simplifying things.
A few days ago, I attempted to verify WIND and WDIR in point_stat using
MADIS observations that were converted using the madis2nc tool you
provided me recently (which works great, by the way). These surface
observations do not contain the U and V components, but instead carry
along WDIR (GC=31) and WIND (GC=32). As you know, the default code
gives you an error saying WDIR is not supported in point_stat if you
happen to request it. Out of curiosity, I commented out the section
that makes the code quit when WDIR is calculated, and used point_stat to
verify WIND/Z10 and WDIR/Z10 against 32/Z10 and 31/Z10, respectively.
The code didn't complain, and the stats file looked generally like I
would expect.
My first question then, is whether this is an acceptable approach when
the obs files contain WIND and WDIR explicitly? One thing I haven't
looked in to yet is whether the code limits the difference of the
observed and forecast wind directions to 180 degrees (as it should), or
whether, for example, an ob/fcst pair of 10/350 would be 340 degrees.
Using non-MADIS observations would obviously be slightly different, but
I would like to try to do that as well. While digging through all the
routines in the vx_met_util directory, it appears that many of the
routines could be used for this purpose. In read_grib.cc, I see that
the routine checks the forecast grib file to see whether U and V are
grid relative, and if so it rotates U and V to earth-relative
coordinates, and also rotates the WDIR field if it's available in the
grib file. The U and V coordinates, relative to earth can then by used
downstream in the convert_u_v_to_wdir routine.
With that info, my second question is, would it be possible to verify
observed WDIR by passing both the observed and forecast U and V
components (which are already earth-relative) into the
convert_u_v_to_wdir routine and compare them directly? It is entirely
possible that I am over-simplifying this in my mind, so please let me
know if this is harder than I'm making it out to be.
The other issue I can think of would be that you would need to verify
wind direction and wind speed together (which makes sense anyway), such
that you can use the observed wind speed threshold for wind direction.
Including the calm observations will create a pretty meaningless wind
direction statistic, so using a wind speed threshold would be necessary.
I would appreciate your comments and/or concerns in this matter, because
it would be wonderful if the community could have a way to verify wind
direction against point observations. We just need to make sure we're
doing it smartly! Thanks again for all your help.
- Jeff Zielonka
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #46651] AutoReply: Wind direction in point_stat
From: Jeff Zielonka
Time: Fri May 13 08:27:41 2011
Hello,
I'm wondering if anyone has had an opportunity to think about this
issue
since this request was made.
As a stop-gap measure to fill some immediate needs, I have found
another
method for verifying wind direction with point_stat outputs that may
be
of interest, either as a way to incorporate the wind direction
statistics calculations directly in MET, or as a post processor or
add-on to the stat_analysis tool.
I am aware that there is a means of calculating wind direction error
with the stat analysis tool, which uses the FBAR and OBAR partial sums
from U and V components. We chose not to use this methodology,
however,
because averaging wind components over an entire domain can result in
a
great deal of cancellation, and therefore wind directions that are
uncharacteristic of what is occurring on the model domain. Instead,
we
wanted to calculate wind direction ob/model pairs at every observation
site and averaged those to get the domain-wide statistic.
In order to accomplish this, we used point_stat to create the MPR
output
file for the U and V components, and wrote a short external program to
read the output and use the pairs to calculate wind direction error
based on the information at each grid point according to a wind speed
threshold. Wind speed are also calculated at each ob site to apply
the
threshold appropriately.
If anyone at DTC would like to look at the code (it's in Fortran, not
C), I would be willing to send it along if you would like to try to
implement it in the MET package somehow. Let me know if there is any
interest in that.
I would still like to know, however, whether it makes sense within the
framework of point_stat to calculate wind direction statistics in a
scalar manner, i.e. WDIR grib vs. WDIR observation. I have noticed
that
the pb2nc tool can derive WDIR from U and V during the conversion
process, and I have confirmed that the resulting directions are
accurate. If this is possible, the use would be able to use wind
directions from standard prepbufr observations as well as from the
output of the madis2nc tool.
Thank you again for your time.
- Jeff Zielonka
On 5/3/2011 1:36 PM, RAL HelpDesk {for met_help at ucar.edu} wrote:
> Greetings,
>
> This message has been automatically generated in response to the
> creation of a trouble ticket regarding:
> "Wind direction in point_stat",
> a summary of which appears below.
>
> There is no need to reply to this message right now. Your ticket
has been
> assigned an ID of [rt.rap.ucar.edu #46651].
>
> Please include the string:
>
> [rt.rap.ucar.edu #46651]
>
> in the subject line of all future correspondence about this issue.
To do so,
> you may reply to this message.
>
> Thank you,
> met_help at ucar.edu
>
>
-------------------------------------------------------------------------
> Hi again,
>
> Let me preface this by saying I realize we've discussed this before
with
> previous versions of MET, but it appears that the new version of
> point_stat (v3.0) is much closer to being able to handle wind
direction
> verification. With that in mind, I wanted to ask a few questions
and
> run a few things by you to make sure I'm not barking up the wrong
tree,
> or over-simplifying things.
>
> A few days ago, I attempted to verify WIND and WDIR in point_stat
using
> MADIS observations that were converted using the madis2nc tool you
> provided me recently (which works great, by the way). These surface
> observations do not contain the U and V components, but instead
carry
> along WDIR (GC=31) and WIND (GC=32). As you know, the default code
> gives you an error saying WDIR is not supported in point_stat if you
> happen to request it. Out of curiosity, I commented out the section
> that makes the code quit when WDIR is calculated, and used
point_stat to
> verify WIND/Z10 and WDIR/Z10 against 32/Z10 and 31/Z10,
respectively.
> The code didn't complain, and the stats file looked generally like I
> would expect.
>
> My first question then, is whether this is an acceptable approach
when
> the obs files contain WIND and WDIR explicitly? One thing I haven't
> looked in to yet is whether the code limits the difference of the
> observed and forecast wind directions to 180 degrees (as it should),
or
> whether, for example, an ob/fcst pair of 10/350 would be 340
degrees.
>
> Using non-MADIS observations would obviously be slightly different,
but
> I would like to try to do that as well. While digging through all
the
> routines in the vx_met_util directory, it appears that many of the
> routines could be used for this purpose. In read_grib.cc, I see
that
> the routine checks the forecast grib file to see whether U and V are
> grid relative, and if so it rotates U and V to earth-relative
> coordinates, and also rotates the WDIR field if it's available in
the
> grib file. The U and V coordinates, relative to earth can then by
used
> downstream in the convert_u_v_to_wdir routine.
>
> With that info, my second question is, would it be possible to
verify
> observed WDIR by passing both the observed and forecast U and V
> components (which are already earth-relative) into the
> convert_u_v_to_wdir routine and compare them directly? It is
entirely
> possible that I am over-simplifying this in my mind, so please let
me
> know if this is harder than I'm making it out to be.
>
> The other issue I can think of would be that you would need to
verify
> wind direction and wind speed together (which makes sense anyway),
such
> that you can use the observed wind speed threshold for wind
direction.
> Including the calm observations will create a pretty meaningless
wind
> direction statistic, so using a wind speed threshold would be
necessary.
>
> I would appreciate your comments and/or concerns in this matter,
because
> it would be wonderful if the community could have a way to verify
wind
> direction against point observations. We just need to make sure
we're
> doing it smartly! Thanks again for all your help.
>
>
> - Jeff Zielonka
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #46651] AutoReply: Wind direction in point_stat
From: John Halley Gotway
Time: Mon May 16 16:03:37 2011
Hi Jeff,
I apologize for the delay in getting back with you. We were very busy
last week standing up a system of real-time verification for a testbed
collaboration project.
I read through your emails regarding the verification of wind
direction. In fact, we discussed this short-coming of MET at a MET
development meeting last week. I too would like to resolve this issue
and get some tools out to the community for verifying wind direction
directly in Grid-Stat and Point-Stat in a statistically appropriate
way. Unfortunately, that's where the trouble comes in. My
understanding is that the statistics we use to assess scalar variables
are not very appropriate for circular variables, like wind speed. The
wrapping from 360 to 0 really throws a wrench into things.
The software engineers working on MET are basically waiting for
scientific direction as to how to implement better support for
verifying wind direction.
I'm going to add Tressa Fowler, a statistician in our group, to this
met_help ticket to get her take on it. Perhaps, she can better
comment on the status.
Thanks,
John Halley Gotway
met_help at ucar.edu
On 05/13/2011 08:27 AM, RAL HelpDesk {for Jeff Zielonka} wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=46651 >
>
> Hello,
>
> I'm wondering if anyone has had an opportunity to think about this
issue
> since this request was made.
>
> As a stop-gap measure to fill some immediate needs, I have found
another
> method for verifying wind direction with point_stat outputs that may
be
> of interest, either as a way to incorporate the wind direction
> statistics calculations directly in MET, or as a post processor or
> add-on to the stat_analysis tool.
>
> I am aware that there is a means of calculating wind direction error
> with the stat analysis tool, which uses the FBAR and OBAR partial
sums
> from U and V components. We chose not to use this methodology,
however,
> because averaging wind components over an entire domain can result
in a
> great deal of cancellation, and therefore wind directions that are
> uncharacteristic of what is occurring on the model domain. Instead,
we
> wanted to calculate wind direction ob/model pairs at every
observation
> site and averaged those to get the domain-wide statistic.
>
> In order to accomplish this, we used point_stat to create the MPR
output
> file for the U and V components, and wrote a short external program
to
> read the output and use the pairs to calculate wind direction error
> based on the information at each grid point according to a wind
speed
> threshold. Wind speed are also calculated at each ob site to apply
the
> threshold appropriately.
>
> If anyone at DTC would like to look at the code (it's in Fortran,
not
> C), I would be willing to send it along if you would like to try to
> implement it in the MET package somehow. Let me know if there is
any
> interest in that.
>
> I would still like to know, however, whether it makes sense within
the
> framework of point_stat to calculate wind direction statistics in a
> scalar manner, i.e. WDIR grib vs. WDIR observation. I have noticed
that
> the pb2nc tool can derive WDIR from U and V during the conversion
> process, and I have confirmed that the resulting directions are
> accurate. If this is possible, the use would be able to use wind
> directions from standard prepbufr observations as well as from the
> output of the madis2nc tool.
>
> Thank you again for your time.
>
>
> - Jeff Zielonka
>
>
> On 5/3/2011 1:36 PM, RAL HelpDesk {for met_help at ucar.edu} wrote:
>> Greetings,
>>
>> This message has been automatically generated in response to the
>> creation of a trouble ticket regarding:
>> "Wind direction in point_stat",
>> a summary of which appears below.
>>
>> There is no need to reply to this message right now. Your ticket
has been
>> assigned an ID of [rt.rap.ucar.edu #46651].
>>
>> Please include the string:
>>
>> [rt.rap.ucar.edu #46651]
>>
>> in the subject line of all future correspondence about this issue.
To do so,
>> you may reply to this message.
>>
>> Thank you,
>> met_help at ucar.edu
>>
>>
-------------------------------------------------------------------------
>> Hi again,
>>
>> Let me preface this by saying I realize we've discussed this before
with
>> previous versions of MET, but it appears that the new version of
>> point_stat (v3.0) is much closer to being able to handle wind
direction
>> verification. With that in mind, I wanted to ask a few questions
and
>> run a few things by you to make sure I'm not barking up the wrong
tree,
>> or over-simplifying things.
>>
>> A few days ago, I attempted to verify WIND and WDIR in point_stat
using
>> MADIS observations that were converted using the madis2nc tool you
>> provided me recently (which works great, by the way). These
surface
>> observations do not contain the U and V components, but instead
carry
>> along WDIR (GC=31) and WIND (GC=32). As you know, the default code
>> gives you an error saying WDIR is not supported in point_stat if
you
>> happen to request it. Out of curiosity, I commented out the
section
>> that makes the code quit when WDIR is calculated, and used
point_stat to
>> verify WIND/Z10 and WDIR/Z10 against 32/Z10 and 31/Z10,
respectively.
>> The code didn't complain, and the stats file looked generally like
I
>> would expect.
>>
>> My first question then, is whether this is an acceptable approach
when
>> the obs files contain WIND and WDIR explicitly? One thing I
haven't
>> looked in to yet is whether the code limits the difference of the
>> observed and forecast wind directions to 180 degrees (as it
should), or
>> whether, for example, an ob/fcst pair of 10/350 would be 340
degrees.
>>
>> Using non-MADIS observations would obviously be slightly different,
but
>> I would like to try to do that as well. While digging through all
the
>> routines in the vx_met_util directory, it appears that many of the
>> routines could be used for this purpose. In read_grib.cc, I see
that
>> the routine checks the forecast grib file to see whether U and V
are
>> grid relative, and if so it rotates U and V to earth-relative
>> coordinates, and also rotates the WDIR field if it's available in
the
>> grib file. The U and V coordinates, relative to earth can then by
used
>> downstream in the convert_u_v_to_wdir routine.
>>
>> With that info, my second question is, would it be possible to
verify
>> observed WDIR by passing both the observed and forecast U and V
>> components (which are already earth-relative) into the
>> convert_u_v_to_wdir routine and compare them directly? It is
entirely
>> possible that I am over-simplifying this in my mind, so please let
me
>> know if this is harder than I'm making it out to be.
>>
>> The other issue I can think of would be that you would need to
verify
>> wind direction and wind speed together (which makes sense anyway),
such
>> that you can use the observed wind speed threshold for wind
direction.
>> Including the calm observations will create a pretty meaningless
wind
>> direction statistic, so using a wind speed threshold would be
necessary.
>>
>> I would appreciate your comments and/or concerns in this matter,
because
>> it would be wonderful if the community could have a way to verify
wind
>> direction against point observations. We just need to make sure
we're
>> doing it smartly! Thanks again for all your help.
>>
>>
>> - Jeff Zielonka
>>
>
------------------------------------------------
More information about the Met_help
mailing list