[Met_help] [rt.rap.ucar.edu #75996] History for MET Input Formats
John Halley Gotway via RT
met_help at ucar.edu
Wed Apr 27 14:17:51 MDT 2016
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
John and company, As I mentioned in previous email, 557th really wants to make use of the HiRA framework proposed by Marion Mittermaier. I plan to use MET as much as possible for this. The first step is to generate the model data neighborhood "probabilities" around each observation point. I have written code to do this and the plan as to convert to NETCdf to feed in to MET. The question is can the FCST file for MET not be a gridded file when read into MET in NETCdf format. If so, then MET could generate the model/ob pairs that end up in MET point stat output. If MET can't handle point sourced FCST data, then I could generate the model/ob pairs myself and then write out the data in a format that stat_analysis can read - maybe.
Thanks
Bob Craig
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Wed Apr 20 12:20:20 2016
Bob,
The point-stat tool reads gridded forecast data and point
observations. So
no, it cannot currently handle point observations.
I'm not very familiar with the Marion's HiRA framework. What are the
final
statistics you'd ultimately like to compute using these neighborhood
probability forecasts and observation pairs? If we can pin down the
statistics, we can figure out what pathway through MET would make the
most
sense.
Thanks,
John
On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> Transaction: Ticket created by robert.craig.2 at us.af.mil
> Queue: met_help
> Subject: MET Input Formats
> Owner: Nobody
> Requestors: robert.craig.2 at us.af.mil
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
>
> John and company, As I mentioned in previous email, 557th really
wants to
> make use of the HiRA framework proposed by Marion Mittermaier. I
plan to
> use MET as much as possible for this. The first step is to
generate the
> model data neighborhood "probabilities" around each observation
point. I
> have written code to do this and the plan as to convert to NETCdf to
feed
> in to MET. The question is can the FCST file for MET not be a
gridded file
> when read into MET in NETCdf format. If so, then MET could generate
the
> model/ob pairs that end up in MET point stat output. If MET can't
handle
> point sourced FCST data, then I could generate the model/ob pairs
myself
> and then write out the data in a format that stat_analysis can read
-
> maybe.
>
> Thanks
> Bob Craig
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Wed Apr 20 12:51:21 2016
Here is the list: for precip accum, total cloud cover, cloud base hgt,
and visibility, calculated the brier skill score from the model/ob
pairs based on the forecasted neighborhoods. The neighborhoods are
defined below:
For each ob, find nearest grid point, get all the grid points in the
neighborhood surrounding this point, if they meet a threshold for the
variable set them to 1, else set to 0. Sum the neighborhood grid
points and divide by the number of points. This gives you a fcst
"probability" to compare with the ob point. The ob is set to 0 or 1
if the event occurred or didn't occur. So you end up with model ob
pairs of the fcst "probability" and the binary ob (typical ensemble
verification).
I have code that will do this. If I write out the results as one of
the MET output types, I was hoping I could then use stat analysis to
compute Brier Skill Scores for these variables.
The rest of the list: temperatures - calculate CRPSS from the
Neighborhood where each grid point in the neighborhood is treated like
an ensemble member. Wind speeds are converted to the Beaufort Scale
(0-9) for the speed in the fcst neighborhoods and same for the
observation. So again, the fcst neighborhood is treated like ensemble
members and a RPSS is calculated.
Supposedly, Barb Brown is familiar with HiRA according to Marion
Mittermaier if you need more detail.
Bob
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Wednesday, April 20, 2016 1:20 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
The point-stat tool reads gridded forecast data and point
observations. So no, it cannot currently handle point observations.
I'm not very familiar with the Marion's HiRA framework. What are the
final statistics you'd ultimately like to compute using these
neighborhood probability forecasts and observation pairs? If we can
pin down the statistics, we can figure out what pathway through MET
would make the most sense.
Thanks,
John
On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> Transaction: Ticket created by robert.craig.2 at us.af.mil
> Queue: met_help
> Subject: MET Input Formats
> Owner: Nobody
> Requestors: robert.craig.2 at us.af.mil
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> >
>
>
> John and company, As I mentioned in previous email, 557th really
wants
> to make use of the HiRA framework proposed by Marion Mittermaier. I
plan to
> use MET as much as possible for this. The first step is to
generate the
> model data neighborhood "probabilities" around each observation
point.
> I have written code to do this and the plan as to convert to NETCdf
to
> feed in to MET. The question is can the FCST file for MET not be a
> gridded file when read into MET in NETCdf format. If so, then MET
> could generate the model/ob pairs that end up in MET point stat
> output. If MET can't handle point sourced FCST data, then I could
> generate the model/ob pairs myself and then write out the data in a
> format that stat_analysis can read - maybe.
>
> Thanks
> Bob Craig
>
>
>
------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Wed Apr 20 13:07:57 2016
Bob,
The first step sounds identical to the computation of a fractional
coverage
fields in Grid-Stat. Grid-Stat computes fractional coverage fields
using a
threshold and neighborhood size as the first step in computing
fractions
skill score (and other neighborhood methods output). Unfortunately,
Grid-Stat does NOT currently write that fractional coverage field out
to a
gridded NetCDF file. Although, a few years ago I hacked it to do so
for
one particular project.
If Grid-Stat was writing those fields out, you could run the field
through
Grid-Stat to compute the gridded probability field and then pass
NetCDF
output of Grid-Stat to Point-Stat. And set up the Point-Stat
configuration
file to interpret that fractional coverage field as a probability.
In the long run, we should enhance Point-Stat to define neighborhood
probabilities like this on the fly. But in the short term, it sure
would
be nice to get the fractional field output from Grid-Stat.
But, as you've mentioned, you have already computed these probability
values. What is the current format of the output? We could probably
write
them to a NetCDF file that MET would understand.
John
On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> Here is the list: for precip accum, total cloud cover, cloud base
hgt, and
> visibility, calculated the brier skill score from the model/ob pairs
based
> on the forecasted neighborhoods. The neighborhoods are defined
below:
>
> For each ob, find nearest grid point, get all the grid points in the
> neighborhood surrounding this point, if they meet a threshold for
the
> variable set them to 1, else set to 0. Sum the neighborhood grid
points
> and divide by the number of points. This gives you a fcst
"probability" to
> compare with the ob point. The ob is set to 0 or 1 if the event
occurred
> or didn't occur. So you end up with model ob pairs of the fcst
> "probability" and the binary ob (typical ensemble verification).
>
> I have code that will do this. If I write out the results as one of
the
> MET output types, I was hoping I could then use stat analysis to
compute
> Brier Skill Scores for these variables.
>
> The rest of the list: temperatures - calculate CRPSS from the
Neighborhood
> where each grid point in the neighborhood is treated like an
ensemble
> member. Wind speeds are converted to the Beaufort Scale (0-9) for
the
> speed in the fcst neighborhoods and same for the observation. So
again,
> the fcst neighborhood is treated like ensemble members and a RPSS is
> calculated.
>
> Supposedly, Barb Brown is familiar with HiRA according to Marion
> Mittermaier if you need more detail.
>
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 1:20 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> The point-stat tool reads gridded forecast data and point
observations.
> So no, it cannot currently handle point observations.
>
> I'm not very familiar with the Marion's HiRA framework. What are
the
> final statistics you'd ultimately like to compute using these
neighborhood
> probability forecasts and observation pairs? If we can pin down the
> statistics, we can figure out what pathway through MET would make
the most
> sense.
>
> Thanks,
> John
>
>
>
> On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > Queue: met_help
> > Subject: MET Input Formats
> > Owner: Nobody
> > Requestors: robert.craig.2 at us.af.mil
> > Status: new
> > Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > >
> >
> >
> > John and company, As I mentioned in previous email, 557th really
wants
> > to make use of the HiRA framework proposed by Marion Mittermaier.
I
> plan to
> > use MET as much as possible for this. The first step is to
generate the
> > model data neighborhood "probabilities" around each observation
point.
> > I have written code to do this and the plan as to convert to
NETCdf to
> > feed in to MET. The question is can the FCST file for MET not be
a
> > gridded file when read into MET in NETCdf format. If so, then MET
> > could generate the model/ob pairs that end up in MET point stat
> > output. If MET can't handle point sourced FCST data, then I could
> > generate the model/ob pairs myself and then write out the data in
a
> > format that stat_analysis can read - maybe.
> >
> > Thanks
> > Bob Craig
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Wed Apr 20 13:31:44 2016
John, I am at the point where I am trying to decide what format to
write the files out to. I could write the fcst neighborhoods to
NetCDF if I can then feed them into MET for point stat. Or I could go
a step further and generate the model/ob pairs then write them out in
a format that MET could use to calculate the stats. Or go a step
further and calculate the statistics and then write them out in a
format the MET stat analysis could read to summarize the data. My
goal is to try to use MET in the greatest extent possible here. Do
you still have your hacked grid_stat code?
BOB
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Wednesday, April 20, 2016 2:08 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
The first step sounds identical to the computation of a fractional
coverage fields in Grid-Stat. Grid-Stat computes fractional coverage
fields using a threshold and neighborhood size as the first step in
computing fractions skill score (and other neighborhood methods
output). Unfortunately, Grid-Stat does NOT currently write that
fractional coverage field out to a gridded NetCDF file. Although, a
few years ago I hacked it to do so for one particular project.
If Grid-Stat was writing those fields out, you could run the field
through Grid-Stat to compute the gridded probability field and then
pass NetCDF output of Grid-Stat to Point-Stat. And set up the Point-
Stat configuration file to interpret that fractional coverage field as
a probability.
In the long run, we should enhance Point-Stat to define neighborhood
probabilities like this on the fly. But in the short term, it sure
would be nice to get the fractional field output from Grid-Stat.
But, as you've mentioned, you have already computed these probability
values. What is the current format of the output? We could probably
write them to a NetCDF file that MET would understand.
John
On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> Here is the list: for precip accum, total cloud cover, cloud base
hgt,
> and visibility, calculated the brier skill score from the model/ob
> pairs based on the forecasted neighborhoods. The neighborhoods are
defined below:
>
> For each ob, find nearest grid point, get all the grid points in the
> neighborhood surrounding this point, if they meet a threshold for
the
> variable set them to 1, else set to 0. Sum the neighborhood grid
> points and divide by the number of points. This gives you a fcst
> "probability" to compare with the ob point. The ob is set to 0 or 1
> if the event occurred or didn't occur. So you end up with model ob
> pairs of the fcst "probability" and the binary ob (typical ensemble
verification).
>
> I have code that will do this. If I write out the results as one of
> the MET output types, I was hoping I could then use stat analysis to
> compute Brier Skill Scores for these variables.
>
> The rest of the list: temperatures - calculate CRPSS from the
> Neighborhood where each grid point in the neighborhood is treated
like
> an ensemble member. Wind speeds are converted to the Beaufort Scale
> (0-9) for the speed in the fcst neighborhoods and same for the
> observation. So again, the fcst neighborhood is treated like
ensemble
> members and a RPSS is calculated.
>
> Supposedly, Barb Brown is familiar with HiRA according to Marion
> Mittermaier if you need more detail.
>
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 1:20 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> The point-stat tool reads gridded forecast data and point
observations.
> So no, it cannot currently handle point observations.
>
> I'm not very familiar with the Marion's HiRA framework. What are
the
> final statistics you'd ultimately like to compute using these
> neighborhood probability forecasts and observation pairs? If we can
> pin down the statistics, we can figure out what pathway through MET
> would make the most sense.
>
> Thanks,
> John
>
>
>
> On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > Queue: met_help
> > Subject: MET Input Formats
> > Owner: Nobody
> > Requestors: robert.craig.2 at us.af.mil
> > Status: new
> > Ticket <URL:
> > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > >
> >
> >
> > John and company, As I mentioned in previous email, 557th really
> > wants to make use of the HiRA framework proposed by Marion
> > Mittermaier. I
> plan to
> > use MET as much as possible for this. The first step is to
generate the
> > model data neighborhood "probabilities" around each observation
point.
> > I have written code to do this and the plan as to convert to
NETCdf
> > to feed in to MET. The question is can the FCST file for MET not
be
> > a gridded file when read into MET in NETCdf format. If so, then
MET
> > could generate the model/ob pairs that end up in MET point stat
> > output. If MET can't handle point sourced FCST data, then I could
> > generate the model/ob pairs myself and then write out the data in
a
> > format that stat_analysis can read - maybe.
> >
> > Thanks
> > Bob Craig
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Wed Apr 20 13:35:41 2016
Bob,
I'd suggest writing them to NetCDF to make them look like the NetCDF
output
from the pcp_combine tool. Yes, Point-Stat can read NetCDF. We've
really
tried to abstract the reading of gridded data out of the tools. As
long as
one MET tool can read a gridded data file, all of them should be able
to.
And you'd be able to plot that NetCDF file using the "plot_data_plane"
utility, which may be helpful.
Thanks,
John
On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, I am at the point where I am trying to decide what format to
write
> the files out to. I could write the fcst neighborhoods to NetCDF if
I can
> then feed them into MET for point stat. Or I could go a step
further and
> generate the model/ob pairs then write them out in a format that MET
could
> use to calculate the stats. Or go a step further and calculate the
> statistics and then write them out in a format the MET stat analysis
could
> read to summarize the data. My goal is to try to use MET in the
greatest
> extent possible here. Do you still have your hacked grid_stat
code?
>
> BOB
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 2:08 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> The first step sounds identical to the computation of a fractional
> coverage fields in Grid-Stat. Grid-Stat computes fractional
coverage
> fields using a threshold and neighborhood size as the first step in
> computing fractions skill score (and other neighborhood methods
output).
> Unfortunately, Grid-Stat does NOT currently write that fractional
coverage
> field out to a gridded NetCDF file. Although, a few years ago I
hacked it
> to do so for one particular project.
>
> If Grid-Stat was writing those fields out, you could run the field
through
> Grid-Stat to compute the gridded probability field and then pass
NetCDF
> output of Grid-Stat to Point-Stat. And set up the Point-Stat
configuration
> file to interpret that fractional coverage field as a probability.
>
> In the long run, we should enhance Point-Stat to define neighborhood
> probabilities like this on the fly. But in the short term, it sure
would
> be nice to get the fractional field output from Grid-Stat.
>
> But, as you've mentioned, you have already computed these
probability
> values. What is the current format of the output? We could
probably write
> them to a NetCDF file that MET would understand.
>
> John
>
> On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > Here is the list: for precip accum, total cloud cover, cloud base
hgt,
> > and visibility, calculated the brier skill score from the model/ob
> > pairs based on the forecasted neighborhoods. The neighborhoods
are
> defined below:
> >
> > For each ob, find nearest grid point, get all the grid points in
the
> > neighborhood surrounding this point, if they meet a threshold for
the
> > variable set them to 1, else set to 0. Sum the neighborhood grid
> > points and divide by the number of points. This gives you a fcst
> > "probability" to compare with the ob point. The ob is set to 0 or
1
> > if the event occurred or didn't occur. So you end up with model
ob
> > pairs of the fcst "probability" and the binary ob (typical
ensemble
> verification).
> >
> > I have code that will do this. If I write out the results as one
of
> > the MET output types, I was hoping I could then use stat analysis
to
> > compute Brier Skill Scores for these variables.
> >
> > The rest of the list: temperatures - calculate CRPSS from the
> > Neighborhood where each grid point in the neighborhood is treated
like
> > an ensemble member. Wind speeds are converted to the Beaufort
Scale
> > (0-9) for the speed in the fcst neighborhoods and same for the
> > observation. So again, the fcst neighborhood is treated like
ensemble
> > members and a RPSS is calculated.
> >
> > Supposedly, Barb Brown is familiar with HiRA according to Marion
> > Mittermaier if you need more detail.
> >
> > Bob
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 1:20 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > The point-stat tool reads gridded forecast data and point
observations.
> > So no, it cannot currently handle point observations.
> >
> > I'm not very familiar with the Marion's HiRA framework. What are
the
> > final statistics you'd ultimately like to compute using these
> > neighborhood probability forecasts and observation pairs? If we
can
> > pin down the statistics, we can figure out what pathway through
MET
> > would make the most sense.
> >
> > Thanks,
> > John
> >
> >
> >
> > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT
<
> > met_help at ucar.edu> wrote:
> >
> > >
> > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > Queue: met_help
> > > Subject: MET Input Formats
> > > Owner: Nobody
> > > Requestors: robert.craig.2 at us.af.mil
> > > Status: new
> > > Ticket <URL:
> > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > >
> > >
> > >
> > > John and company, As I mentioned in previous email, 557th really
> > > wants to make use of the HiRA framework proposed by Marion
> > > Mittermaier. I
> > plan to
> > > use MET as much as possible for this. The first step is to
generate
> the
> > > model data neighborhood "probabilities" around each observation
point.
> > > I have written code to do this and the plan as to convert to
NETCdf
> > > to feed in to MET. The question is can the FCST file for MET
not be
> > > a gridded file when read into MET in NETCdf format. If so, then
MET
> > > could generate the model/ob pairs that end up in MET point stat
> > > output. If MET can't handle point sourced FCST data, then I
could
> > > generate the model/ob pairs myself and then write out the data
in a
> > > format that stat_analysis can read - maybe.
> > >
> > > Thanks
> > > Bob Craig
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Wed Apr 20 13:45:54 2016
Okay, so I will write them out in NETCDF to look like output from the
pcp_combine tool. In point stat, this data will considered the
forecast data. Prep buffer will be the obs data. This would be
great, since I can use MET for the stat calculations.
Thanks
Bob
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Wednesday, April 20, 2016 2:36 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
I'd suggest writing them to NetCDF to make them look like the NetCDF
output from the pcp_combine tool. Yes, Point-Stat can read NetCDF.
We've really tried to abstract the reading of gridded data out of the
tools. As long as one MET tool can read a gridded data file, all of
them should be able to.
And you'd be able to plot that NetCDF file using the "plot_data_plane"
utility, which may be helpful.
Thanks,
John
On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, I am at the point where I am trying to decide what format to
> write the files out to. I could write the fcst neighborhoods to
> NetCDF if I can then feed them into MET for point stat. Or I could
go
> a step further and generate the model/ob pairs then write them out
in
> a format that MET could use to calculate the stats. Or go a step
> further and calculate the statistics and then write them out in a
> format the MET stat analysis could read to summarize the data. My
goal is to try to use MET in the greatest
> extent possible here. Do you still have your hacked grid_stat
code?
>
> BOB
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 2:08 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> The first step sounds identical to the computation of a fractional
> coverage fields in Grid-Stat. Grid-Stat computes fractional
coverage
> fields using a threshold and neighborhood size as the first step in
> computing fractions skill score (and other neighborhood methods
output).
> Unfortunately, Grid-Stat does NOT currently write that fractional
> coverage field out to a gridded NetCDF file. Although, a few years
> ago I hacked it to do so for one particular project.
>
> If Grid-Stat was writing those fields out, you could run the field
> through Grid-Stat to compute the gridded probability field and then
> pass NetCDF output of Grid-Stat to Point-Stat. And set up the
> Point-Stat configuration file to interpret that fractional coverage
field as a probability.
>
> In the long run, we should enhance Point-Stat to define neighborhood
> probabilities like this on the fly. But in the short term, it sure
> would be nice to get the fractional field output from Grid-Stat.
>
> But, as you've mentioned, you have already computed these
probability
> values. What is the current format of the output? We could
probably
> write them to a NetCDF file that MET would understand.
>
> John
>
> On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > Here is the list: for precip accum, total cloud cover, cloud base
> > hgt, and visibility, calculated the brier skill score from the
> > model/ob pairs based on the forecasted neighborhoods. The
> > neighborhoods are
> defined below:
> >
> > For each ob, find nearest grid point, get all the grid points in
the
> > neighborhood surrounding this point, if they meet a threshold for
> > the variable set them to 1, else set to 0. Sum the neighborhood
> > grid points and divide by the number of points. This gives you a
> > fcst "probability" to compare with the ob point. The ob is set to
0
> > or 1 if the event occurred or didn't occur. So you end up with
> > model ob pairs of the fcst "probability" and the binary ob
(typical
> > ensemble
> verification).
> >
> > I have code that will do this. If I write out the results as one
of
> > the MET output types, I was hoping I could then use stat analysis
to
> > compute Brier Skill Scores for these variables.
> >
> > The rest of the list: temperatures - calculate CRPSS from the
> > Neighborhood where each grid point in the neighborhood is treated
> > like an ensemble member. Wind speeds are converted to the
Beaufort
> > Scale
> > (0-9) for the speed in the fcst neighborhoods and same for the
> > observation. So again, the fcst neighborhood is treated like
> > ensemble members and a RPSS is calculated.
> >
> > Supposedly, Barb Brown is familiar with HiRA according to Marion
> > Mittermaier if you need more detail.
> >
> > Bob
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 1:20 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > The point-stat tool reads gridded forecast data and point
observations.
> > So no, it cannot currently handle point observations.
> >
> > I'm not very familiar with the Marion's HiRA framework. What are
> > the final statistics you'd ultimately like to compute using these
> > neighborhood probability forecasts and observation pairs? If we
can
> > pin down the statistics, we can figure out what pathway through
MET
> > would make the most sense.
> >
> > Thanks,
> > John
> >
> >
> >
> > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via RT
<
> > met_help at ucar.edu> wrote:
> >
> > >
> > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > Queue: met_help
> > > Subject: MET Input Formats
> > > Owner: Nobody
> > > Requestors: robert.craig.2 at us.af.mil
> > > Status: new
> > > Ticket <URL:
> > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > >
> > >
> > >
> > > John and company, As I mentioned in previous email, 557th really
> > > wants to make use of the HiRA framework proposed by Marion
> > > Mittermaier. I
> > plan to
> > > use MET as much as possible for this. The first step is to
generate
> the
> > > model data neighborhood "probabilities" around each observation
point.
> > > I have written code to do this and the plan as to convert to
> > > NETCdf to feed in to MET. The question is can the FCST file for
> > > MET not be a gridded file when read into MET in NETCdf format.
If
> > > so, then MET could generate the model/ob pairs that end up in
MET
> > > point stat output. If MET can't handle point sourced FCST data,
> > > then I could generate the model/ob pairs myself and then write
out
> > > the data in a format that stat_analysis can read - maybe.
> > >
> > > Thanks
> > > Bob Craig
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Wed Apr 20 14:09:33 2016
Bob,
That sounds good to me. Once you have a sample NetCDF output file,
feel
free to send it to me. I can run it through Point-Stat and send you
suggested configuration file to use.
Thanks,
John
On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> Okay, so I will write them out in NETCDF to look like output from
the
> pcp_combine tool. In point stat, this data will considered the
forecast
> data. Prep buffer will be the obs data. This would be great, since
I can
> use MET for the stat calculations.
>
> Thanks
>
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 2:36 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> I'd suggest writing them to NetCDF to make them look like the NetCDF
> output from the pcp_combine tool. Yes, Point-Stat can read NetCDF.
We've
> really tried to abstract the reading of gridded data out of the
tools. As
> long as one MET tool can read a gridded data file, all of them
should be
> able to.
>
> And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> utility, which may be helpful.
>
> Thanks,
> John
>
> On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > John, I am at the point where I am trying to decide what format to
> > write the files out to. I could write the fcst neighborhoods to
> > NetCDF if I can then feed them into MET for point stat. Or I
could go
> > a step further and generate the model/ob pairs then write them out
in
> > a format that MET could use to calculate the stats. Or go a step
> > further and calculate the statistics and then write them out in a
> > format the MET stat analysis could read to summarize the data. My
goal
> is to try to use MET in the greatest
> > extent possible here. Do you still have your hacked grid_stat
code?
> >
> > BOB
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 2:08 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > The first step sounds identical to the computation of a fractional
> > coverage fields in Grid-Stat. Grid-Stat computes fractional
coverage
> > fields using a threshold and neighborhood size as the first step
in
> > computing fractions skill score (and other neighborhood methods
output).
> > Unfortunately, Grid-Stat does NOT currently write that fractional
> > coverage field out to a gridded NetCDF file. Although, a few
years
> > ago I hacked it to do so for one particular project.
> >
> > If Grid-Stat was writing those fields out, you could run the field
> > through Grid-Stat to compute the gridded probability field and
then
> > pass NetCDF output of Grid-Stat to Point-Stat. And set up the
> > Point-Stat configuration file to interpret that fractional
coverage
> field as a probability.
> >
> > In the long run, we should enhance Point-Stat to define
neighborhood
> > probabilities like this on the fly. But in the short term, it
sure
> > would be nice to get the fractional field output from Grid-Stat.
> >
> > But, as you've mentioned, you have already computed these
probability
> > values. What is the current format of the output? We could
probably
> > write them to a NetCDF file that MET would understand.
> >
> > John
> >
> > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT
<
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > Here is the list: for precip accum, total cloud cover, cloud
base
> > > hgt, and visibility, calculated the brier skill score from the
> > > model/ob pairs based on the forecasted neighborhoods. The
> > > neighborhoods are
> > defined below:
> > >
> > > For each ob, find nearest grid point, get all the grid points in
the
> > > neighborhood surrounding this point, if they meet a threshold
for
> > > the variable set them to 1, else set to 0. Sum the neighborhood
> > > grid points and divide by the number of points. This gives you
a
> > > fcst "probability" to compare with the ob point. The ob is set
to 0
> > > or 1 if the event occurred or didn't occur. So you end up with
> > > model ob pairs of the fcst "probability" and the binary ob
(typical
> > > ensemble
> > verification).
> > >
> > > I have code that will do this. If I write out the results as
one of
> > > the MET output types, I was hoping I could then use stat
analysis to
> > > compute Brier Skill Scores for these variables.
> > >
> > > The rest of the list: temperatures - calculate CRPSS from the
> > > Neighborhood where each grid point in the neighborhood is
treated
> > > like an ensemble member. Wind speeds are converted to the
Beaufort
> > > Scale
> > > (0-9) for the speed in the fcst neighborhoods and same for the
> > > observation. So again, the fcst neighborhood is treated like
> > > ensemble members and a RPSS is calculated.
> > >
> > > Supposedly, Barb Brown is familiar with HiRA according to Marion
> > > Mittermaier if you need more detail.
> > >
> > > Bob
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > The point-stat tool reads gridded forecast data and point
observations.
> > > So no, it cannot currently handle point observations.
> > >
> > > I'm not very familiar with the Marion's HiRA framework. What
are
> > > the final statistics you'd ultimately like to compute using
these
> > > neighborhood probability forecasts and observation pairs? If we
can
> > > pin down the statistics, we can figure out what pathway through
MET
> > > would make the most sense.
> > >
> > > Thanks,
> > > John
> > >
> > >
> > >
> > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via
RT <
> > > met_help at ucar.edu> wrote:
> > >
> > > >
> > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > Queue: met_help
> > > > Subject: MET Input Formats
> > > > Owner: Nobody
> > > > Requestors: robert.craig.2 at us.af.mil
> > > > Status: new
> > > > Ticket <URL:
> > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > >
> > > >
> > > >
> > > > John and company, As I mentioned in previous email, 557th
really
> > > > wants to make use of the HiRA framework proposed by Marion
> > > > Mittermaier. I
> > > plan to
> > > > use MET as much as possible for this. The first step is to
generate
> > the
> > > > model data neighborhood "probabilities" around each
observation
> point.
> > > > I have written code to do this and the plan as to convert to
> > > > NETCdf to feed in to MET. The question is can the FCST file
for
> > > > MET not be a gridded file when read into MET in NETCdf format.
If
> > > > so, then MET could generate the model/ob pairs that end up in
MET
> > > > point stat output. If MET can't handle point sourced FCST
data,
> > > > then I could generate the model/ob pairs myself and then write
out
> > > > the data in a format that stat_analysis can read - maybe.
> > > >
> > > > Thanks
> > > > Bob Craig
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Wed Apr 20 14:21:01 2016
Okay, sounds good.
Thanks
Bob
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Wednesday, April 20, 2016 3:10 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
That sounds good to me. Once you have a sample NetCDF output file,
feel free to send it to me. I can run it through Point-Stat and send
you suggested configuration file to use.
Thanks,
John
On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> Okay, so I will write them out in NETCDF to look like output from
the
> pcp_combine tool. In point stat, this data will considered the
> forecast data. Prep buffer will be the obs data. This would be
> great, since I can use MET for the stat calculations.
>
> Thanks
>
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 2:36 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> I'd suggest writing them to NetCDF to make them look like the NetCDF
> output from the pcp_combine tool. Yes, Point-Stat can read NetCDF.
> We've really tried to abstract the reading of gridded data out of
the
> tools. As long as one MET tool can read a gridded data file, all of
> them should be able to.
>
> And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> utility, which may be helpful.
>
> Thanks,
> John
>
> On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > John, I am at the point where I am trying to decide what format to
> > write the files out to. I could write the fcst neighborhoods to
> > NetCDF if I can then feed them into MET for point stat. Or I
could
> > go a step further and generate the model/ob pairs then write them
> > out in a format that MET could use to calculate the stats. Or go
a
> > step further and calculate the statistics and then write them out
in
> > a format the MET stat analysis could read to summarize the data.
My
> > goal
> is to try to use MET in the greatest
> > extent possible here. Do you still have your hacked grid_stat
code?
> >
> > BOB
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 2:08 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > The first step sounds identical to the computation of a fractional
> > coverage fields in Grid-Stat. Grid-Stat computes fractional
> > coverage fields using a threshold and neighborhood size as the
first
> > step in computing fractions skill score (and other neighborhood
methods output).
> > Unfortunately, Grid-Stat does NOT currently write that fractional
> > coverage field out to a gridded NetCDF file. Although, a few
years
> > ago I hacked it to do so for one particular project.
> >
> > If Grid-Stat was writing those fields out, you could run the field
> > through Grid-Stat to compute the gridded probability field and
then
> > pass NetCDF output of Grid-Stat to Point-Stat. And set up the
> > Point-Stat configuration file to interpret that fractional
coverage
> field as a probability.
> >
> > In the long run, we should enhance Point-Stat to define
neighborhood
> > probabilities like this on the fly. But in the short term, it
sure
> > would be nice to get the fractional field output from Grid-Stat.
> >
> > But, as you've mentioned, you have already computed these
> > probability values. What is the current format of the output? We
> > could probably write them to a NetCDF file that MET would
understand.
> >
> > John
> >
> > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT
<
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > Here is the list: for precip accum, total cloud cover, cloud
base
> > > hgt, and visibility, calculated the brier skill score from the
> > > model/ob pairs based on the forecasted neighborhoods. The
> > > neighborhoods are
> > defined below:
> > >
> > > For each ob, find nearest grid point, get all the grid points in
> > > the neighborhood surrounding this point, if they meet a
threshold
> > > for the variable set them to 1, else set to 0. Sum the
> > > neighborhood grid points and divide by the number of points.
This
> > > gives you a fcst "probability" to compare with the ob point.
The
> > > ob is set to 0 or 1 if the event occurred or didn't occur. So
you
> > > end up with model ob pairs of the fcst "probability" and the
> > > binary ob (typical ensemble
> > verification).
> > >
> > > I have code that will do this. If I write out the results as
one
> > > of the MET output types, I was hoping I could then use stat
> > > analysis to compute Brier Skill Scores for these variables.
> > >
> > > The rest of the list: temperatures - calculate CRPSS from the
> > > Neighborhood where each grid point in the neighborhood is
treated
> > > like an ensemble member. Wind speeds are converted to the
> > > Beaufort Scale
> > > (0-9) for the speed in the fcst neighborhoods and same for the
> > > observation. So again, the fcst neighborhood is treated like
> > > ensemble members and a RPSS is calculated.
> > >
> > > Supposedly, Barb Brown is familiar with HiRA according to Marion
> > > Mittermaier if you need more detail.
> > >
> > > Bob
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > The point-stat tool reads gridded forecast data and point
observations.
> > > So no, it cannot currently handle point observations.
> > >
> > > I'm not very familiar with the Marion's HiRA framework. What
are
> > > the final statistics you'd ultimately like to compute using
these
> > > neighborhood probability forecasts and observation pairs? If we
> > > can pin down the statistics, we can figure out what pathway
> > > through MET would make the most sense.
> > >
> > > Thanks,
> > > John
> > >
> > >
> > >
> > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via
RT
> > > < met_help at ucar.edu> wrote:
> > >
> > > >
> > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > Queue: met_help
> > > > Subject: MET Input Formats
> > > > Owner: Nobody
> > > > Requestors: robert.craig.2 at us.af.mil
> > > > Status: new
> > > > Ticket <URL:
> > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > >
> > > >
> > > >
> > > > John and company, As I mentioned in previous email, 557th
really
> > > > wants to make use of the HiRA framework proposed by Marion
> > > > Mittermaier. I
> > > plan to
> > > > use MET as much as possible for this. The first step is to
generate
> > the
> > > > model data neighborhood "probabilities" around each
observation
> point.
> > > > I have written code to do this and the plan as to convert to
> > > > NETCdf to feed in to MET. The question is can the FCST file
for
> > > > MET not be a gridded file when read into MET in NETCdf format.
> > > > If so, then MET could generate the model/ob pairs that end up
in
> > > > MET point stat output. If MET can't handle point sourced FCST
> > > > data, then I could generate the model/ob pairs myself and then
> > > > write out the data in a format that stat_analysis can read -
maybe.
> > > >
> > > > Thanks
> > > > Bob Craig
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Fri Apr 22 09:58:52 2016
John, I am currently coding up the NETCDF file. I am not sure what
global attributes I need to define for point data. Is there a
reference on what attributes MET expects in a NETCDF file for
different types of input data?
Thanks
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Wednesday, April 20, 2016 3:10 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
That sounds good to me. Once you have a sample NetCDF output file,
feel free to send it to me. I can run it through Point-Stat and send
you suggested configuration file to use.
Thanks,
John
On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> Okay, so I will write them out in NETCDF to look like output from
the
> pcp_combine tool. In point stat, this data will considered the
> forecast data. Prep buffer will be the obs data. This would be
> great, since I can use MET for the stat calculations.
>
> Thanks
>
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 2:36 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> I'd suggest writing them to NetCDF to make them look like the NetCDF
> output from the pcp_combine tool. Yes, Point-Stat can read NetCDF.
> We've really tried to abstract the reading of gridded data out of
the
> tools. As long as one MET tool can read a gridded data file, all of
> them should be able to.
>
> And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> utility, which may be helpful.
>
> Thanks,
> John
>
> On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > John, I am at the point where I am trying to decide what format to
> > write the files out to. I could write the fcst neighborhoods to
> > NetCDF if I can then feed them into MET for point stat. Or I
could
> > go a step further and generate the model/ob pairs then write them
> > out in a format that MET could use to calculate the stats. Or go
a
> > step further and calculate the statistics and then write them out
in
> > a format the MET stat analysis could read to summarize the data.
My
> > goal
> is to try to use MET in the greatest
> > extent possible here. Do you still have your hacked grid_stat
code?
> >
> > BOB
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 2:08 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > The first step sounds identical to the computation of a fractional
> > coverage fields in Grid-Stat. Grid-Stat computes fractional
> > coverage fields using a threshold and neighborhood size as the
first
> > step in computing fractions skill score (and other neighborhood
methods output).
> > Unfortunately, Grid-Stat does NOT currently write that fractional
> > coverage field out to a gridded NetCDF file. Although, a few
years
> > ago I hacked it to do so for one particular project.
> >
> > If Grid-Stat was writing those fields out, you could run the field
> > through Grid-Stat to compute the gridded probability field and
then
> > pass NetCDF output of Grid-Stat to Point-Stat. And set up the
> > Point-Stat configuration file to interpret that fractional
coverage
> field as a probability.
> >
> > In the long run, we should enhance Point-Stat to define
neighborhood
> > probabilities like this on the fly. But in the short term, it
sure
> > would be nice to get the fractional field output from Grid-Stat.
> >
> > But, as you've mentioned, you have already computed these
> > probability values. What is the current format of the output? We
> > could probably write them to a NetCDF file that MET would
understand.
> >
> > John
> >
> > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via RT
<
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > Here is the list: for precip accum, total cloud cover, cloud
base
> > > hgt, and visibility, calculated the brier skill score from the
> > > model/ob pairs based on the forecasted neighborhoods. The
> > > neighborhoods are
> > defined below:
> > >
> > > For each ob, find nearest grid point, get all the grid points in
> > > the neighborhood surrounding this point, if they meet a
threshold
> > > for the variable set them to 1, else set to 0. Sum the
> > > neighborhood grid points and divide by the number of points.
This
> > > gives you a fcst "probability" to compare with the ob point.
The
> > > ob is set to 0 or 1 if the event occurred or didn't occur. So
you
> > > end up with model ob pairs of the fcst "probability" and the
> > > binary ob (typical ensemble
> > verification).
> > >
> > > I have code that will do this. If I write out the results as
one
> > > of the MET output types, I was hoping I could then use stat
> > > analysis to compute Brier Skill Scores for these variables.
> > >
> > > The rest of the list: temperatures - calculate CRPSS from the
> > > Neighborhood where each grid point in the neighborhood is
treated
> > > like an ensemble member. Wind speeds are converted to the
> > > Beaufort Scale
> > > (0-9) for the speed in the fcst neighborhoods and same for the
> > > observation. So again, the fcst neighborhood is treated like
> > > ensemble members and a RPSS is calculated.
> > >
> > > Supposedly, Barb Brown is familiar with HiRA according to Marion
> > > Mittermaier if you need more detail.
> > >
> > > Bob
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > The point-stat tool reads gridded forecast data and point
observations.
> > > So no, it cannot currently handle point observations.
> > >
> > > I'm not very familiar with the Marion's HiRA framework. What
are
> > > the final statistics you'd ultimately like to compute using
these
> > > neighborhood probability forecasts and observation pairs? If we
> > > can pin down the statistics, we can figure out what pathway
> > > through MET would make the most sense.
> > >
> > > Thanks,
> > > John
> > >
> > >
> > >
> > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via
RT
> > > < met_help at ucar.edu> wrote:
> > >
> > > >
> > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > Queue: met_help
> > > > Subject: MET Input Formats
> > > > Owner: Nobody
> > > > Requestors: robert.craig.2 at us.af.mil
> > > > Status: new
> > > > Ticket <URL:
> > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > >
> > > >
> > > >
> > > > John and company, As I mentioned in previous email, 557th
really
> > > > wants to make use of the HiRA framework proposed by Marion
> > > > Mittermaier. I
> > > plan to
> > > > use MET as much as possible for this. The first step is to
generate
> > the
> > > > model data neighborhood "probabilities" around each
observation
> point.
> > > > I have written code to do this and the plan as to convert to
> > > > NETCdf to feed in to MET. The question is can the FCST file
for
> > > > MET not be a gridded file when read into MET in NETCdf format.
> > > > If so, then MET could generate the model/ob pairs that end up
in
> > > > MET point stat output. If MET can't handle point sourced FCST
> > > > data, then I could generate the model/ob pairs myself and then
> > > > write out the data in a format that stat_analysis can read -
maybe.
> > > >
> > > > Thanks
> > > > Bob Craig
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Fri Apr 22 15:27:29 2016
Bob,
Unfortunately no, we don't have anything well-documented. Eventually,
we
plan to switch from this very limited format to using CF-compliant
NetCDF.
I'll describe the current format for you below. This is basically the
same
format at the output of the pcp_combine tool (and all of the other
NetCDF
output files MET writes). I've attached an example NetCDF header from
pcp_combine. Here are the relevant points:
(1) Must define dimensions named "lat" and "lon" that match the grid
dimensions.
(2) You do *NOT* have to define corresponding variables for "lat" and
"lon"
but you can if you'd like.
(3) The data variable (e.g. "APCP_12" in the attached example) must
have
dimensions (lat, lon).
(4) Note that it's fine to define multiple variables and give them
different names if you'd like.
(5) Note that you can name the variables anything you'd like.
(6) Specify variable attributes strings for: name, long_name, level,
and
units.
(7) Specify _FillValue attribute as a float value of -9999.
(8) Specify timing info as strings: init_time_ut, valid_time_ut, and
accum_time_sec.
We store these as strings instead of integers to avoid overflow
problems in NetCDF.
FYI, you can specify strings for init_time, valid_time, and
accum_time
if you'd like, but MET does *NOT* actually read them.
(9) In the global attributes, define: MET_version and Projection
information (that's all the attributes after Projection and down to
ny).
To see the NetCDF projection definition for your data, just run one of
your
GRIB files through pcp_combine and look at the output. Use the same
projection settings in the NetCDF file you're creating.
If you run into trouble, just let me know.
Thanks,
John
On Fri, Apr 22, 2016 at 9:58 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, I am currently coding up the NETCDF file. I am not sure what
global
> attributes I need to define for point data. Is there a reference on
what
> attributes MET expects in a NETCDF file for different types of input
data?
>
> Thanks
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 3:10 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> That sounds good to me. Once you have a sample NetCDF output file,
feel
> free to send it to me. I can run it through Point-Stat and send you
> suggested configuration file to use.
>
> Thanks,
> John
>
> On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > Okay, so I will write them out in NETCDF to look like output from
the
> > pcp_combine tool. In point stat, this data will considered the
> > forecast data. Prep buffer will be the obs data. This would be
> > great, since I can use MET for the stat calculations.
> >
> > Thanks
> >
> > Bob
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 2:36 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > I'd suggest writing them to NetCDF to make them look like the
NetCDF
> > output from the pcp_combine tool. Yes, Point-Stat can read
NetCDF.
> > We've really tried to abstract the reading of gridded data out of
the
> > tools. As long as one MET tool can read a gridded data file, all
of
> > them should be able to.
> >
> > And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> > utility, which may be helpful.
> >
> > Thanks,
> > John
> >
> > On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > John, I am at the point where I am trying to decide what format
to
> > > write the files out to. I could write the fcst neighborhoods to
> > > NetCDF if I can then feed them into MET for point stat. Or I
could
> > > go a step further and generate the model/ob pairs then write
them
> > > out in a format that MET could use to calculate the stats. Or
go a
> > > step further and calculate the statistics and then write them
out in
> > > a format the MET stat analysis could read to summarize the data.
My
> > > goal
> > is to try to use MET in the greatest
> > > extent possible here. Do you still have your hacked
grid_stat code?
> > >
> > > BOB
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 2:08 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > The first step sounds identical to the computation of a
fractional
> > > coverage fields in Grid-Stat. Grid-Stat computes fractional
> > > coverage fields using a threshold and neighborhood size as the
first
> > > step in computing fractions skill score (and other neighborhood
> methods output).
> > > Unfortunately, Grid-Stat does NOT currently write that
fractional
> > > coverage field out to a gridded NetCDF file. Although, a few
years
> > > ago I hacked it to do so for one particular project.
> > >
> > > If Grid-Stat was writing those fields out, you could run the
field
> > > through Grid-Stat to compute the gridded probability field and
then
> > > pass NetCDF output of Grid-Stat to Point-Stat. And set up the
> > > Point-Stat configuration file to interpret that fractional
coverage
> > field as a probability.
> > >
> > > In the long run, we should enhance Point-Stat to define
neighborhood
> > > probabilities like this on the fly. But in the short term, it
sure
> > > would be nice to get the fractional field output from Grid-Stat.
> > >
> > > But, as you've mentioned, you have already computed these
> > > probability values. What is the current format of the output?
We
> > > could probably write them to a NetCDF file that MET would
understand.
> > >
> > > John
> > >
> > > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via
RT <
> > > met_help at ucar.edu> wrote:
> > >
> > > >
> > > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
>
> > > >
> > > > Here is the list: for precip accum, total cloud cover, cloud
base
> > > > hgt, and visibility, calculated the brier skill score from the
> > > > model/ob pairs based on the forecasted neighborhoods. The
> > > > neighborhoods are
> > > defined below:
> > > >
> > > > For each ob, find nearest grid point, get all the grid points
in
> > > > the neighborhood surrounding this point, if they meet a
threshold
> > > > for the variable set them to 1, else set to 0. Sum the
> > > > neighborhood grid points and divide by the number of points.
This
> > > > gives you a fcst "probability" to compare with the ob point.
The
> > > > ob is set to 0 or 1 if the event occurred or didn't occur. So
you
> > > > end up with model ob pairs of the fcst "probability" and the
> > > > binary ob (typical ensemble
> > > verification).
> > > >
> > > > I have code that will do this. If I write out the results as
one
> > > > of the MET output types, I was hoping I could then use stat
> > > > analysis to compute Brier Skill Scores for these variables.
> > > >
> > > > The rest of the list: temperatures - calculate CRPSS from the
> > > > Neighborhood where each grid point in the neighborhood is
treated
> > > > like an ensemble member. Wind speeds are converted to the
> > > > Beaufort Scale
> > > > (0-9) for the speed in the fcst neighborhoods and same for the
> > > > observation. So again, the fcst neighborhood is treated like
> > > > ensemble members and a RPSS is calculated.
> > > >
> > > > Supposedly, Barb Brown is familiar with HiRA according to
Marion
> > > > Mittermaier if you need more detail.
> > > >
> > > > Bob
> > > >
> > > > -----Original Message-----
> > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > <robert.craig.2 at us.af.mil>
> > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > >
> > > > Bob,
> > > >
> > > > The point-stat tool reads gridded forecast data and point
> observations.
> > > > So no, it cannot currently handle point observations.
> > > >
> > > > I'm not very familiar with the Marion's HiRA framework. What
are
> > > > the final statistics you'd ultimately like to compute using
these
> > > > neighborhood probability forecasts and observation pairs? If
we
> > > > can pin down the statistics, we can figure out what pathway
> > > > through MET would make the most sense.
> > > >
> > > > Thanks,
> > > > John
> > > >
> > > >
> > > >
> > > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via
RT
> > > > < met_help at ucar.edu> wrote:
> > > >
> > > > >
> > > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > > Queue: met_help
> > > > > Subject: MET Input Formats
> > > > > Owner: Nobody
> > > > > Requestors: robert.craig.2 at us.af.mil
> > > > > Status: new
> > > > > Ticket <URL:
> > > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > > >
> > > > >
> > > > >
> > > > > John and company, As I mentioned in previous email, 557th
really
> > > > > wants to make use of the HiRA framework proposed by Marion
> > > > > Mittermaier. I
> > > > plan to
> > > > > use MET as much as possible for this. The first step is to
> generate
> > > the
> > > > > model data neighborhood "probabilities" around each
observation
> > point.
> > > > > I have written code to do this and the plan as to convert to
> > > > > NETCdf to feed in to MET. The question is can the FCST file
for
> > > > > MET not be a gridded file when read into MET in NETCdf
format.
> > > > > If so, then MET could generate the model/ob pairs that end
up in
> > > > > MET point stat output. If MET can't handle point sourced
FCST
> > > > > data, then I could generate the model/ob pairs myself and
then
> > > > > write out the data in a format that stat_analysis can read
-
> maybe.
> > > > >
> > > > > Thanks
> > > > > Bob Craig
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Mon Apr 25 06:51:26 2016
John, the key question is the data file is not going to be gridded.
It will contain the model neighborhood values for each ob site. For
the HiRA framework, we compare the model neighborhood value calculated
at each obs site and compare to the ob value. I was hoping to use
pointstat for this but I was not sure of the NETCDF header parameters
for a non-gridded (or 1 dimensional grid) data set.
Thanks
Bob
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Friday, April 22, 2016 4:27 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
Unfortunately no, we don't have anything well-documented. Eventually,
we plan to switch from this very limited format to using CF-compliant
NetCDF.
I'll describe the current format for you below. This is basically the
same format at the output of the pcp_combine tool (and all of the
other NetCDF output files MET writes). I've attached an example
NetCDF header from pcp_combine. Here are the relevant points:
(1) Must define dimensions named "lat" and "lon" that match the grid
dimensions.
(2) You do *NOT* have to define corresponding variables for "lat" and
"lon"
but you can if you'd like.
(3) The data variable (e.g. "APCP_12" in the attached example) must
have dimensions (lat, lon).
(4) Note that it's fine to define multiple variables and give them
different names if you'd like.
(5) Note that you can name the variables anything you'd like.
(6) Specify variable attributes strings for: name, long_name, level,
and units.
(7) Specify _FillValue attribute as a float value of -9999.
(8) Specify timing info as strings: init_time_ut, valid_time_ut, and
accum_time_sec.
We store these as strings instead of integers to avoid overflow
problems in NetCDF.
FYI, you can specify strings for init_time, valid_time, and
accum_time if you'd like, but MET does *NOT* actually read them.
(9) In the global attributes, define: MET_version and Projection
information (that's all the attributes after Projection and down to
ny).
To see the NetCDF projection definition for your data, just run one of
your GRIB files through pcp_combine and look at the output. Use the
same projection settings in the NetCDF file you're creating.
If you run into trouble, just let me know.
Thanks,
John
On Fri, Apr 22, 2016 at 9:58 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, I am currently coding up the NETCDF file. I am not sure what
> global attributes I need to define for point data. Is there a
> reference on what attributes MET expects in a NETCDF file for
different types of input data?
>
> Thanks
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Wednesday, April 20, 2016 3:10 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> That sounds good to me. Once you have a sample NetCDF output file,
> feel free to send it to me. I can run it through Point-Stat and
send
> you suggested configuration file to use.
>
> Thanks,
> John
>
> On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > Okay, so I will write them out in NETCDF to look like output from
> > the pcp_combine tool. In point stat, this data will considered
the
> > forecast data. Prep buffer will be the obs data. This would be
> > great, since I can use MET for the stat calculations.
> >
> > Thanks
> >
> > Bob
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 2:36 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > I'd suggest writing them to NetCDF to make them look like the
NetCDF
> > output from the pcp_combine tool. Yes, Point-Stat can read
NetCDF.
> > We've really tried to abstract the reading of gridded data out of
> > the tools. As long as one MET tool can read a gridded data file,
> > all of them should be able to.
> >
> > And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> > utility, which may be helpful.
> >
> > Thanks,
> > John
> >
> > On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT <
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > John, I am at the point where I am trying to decide what format
to
> > > write the files out to. I could write the fcst neighborhoods to
> > > NetCDF if I can then feed them into MET for point stat. Or I
> > > could go a step further and generate the model/ob pairs then
write
> > > them out in a format that MET could use to calculate the stats.
> > > Or go a step further and calculate the statistics and then write
> > > them out in a format the MET stat analysis could read to
summarize
> > > the data. My goal
> > is to try to use MET in the greatest
> > > extent possible here. Do you still have your hacked
grid_stat code?
> > >
> > > BOB
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 2:08 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > The first step sounds identical to the computation of a
fractional
> > > coverage fields in Grid-Stat. Grid-Stat computes fractional
> > > coverage fields using a threshold and neighborhood size as the
> > > first step in computing fractions skill score (and other
> > > neighborhood
> methods output).
> > > Unfortunately, Grid-Stat does NOT currently write that
fractional
> > > coverage field out to a gridded NetCDF file. Although, a few
> > > years ago I hacked it to do so for one particular project.
> > >
> > > If Grid-Stat was writing those fields out, you could run the
field
> > > through Grid-Stat to compute the gridded probability field and
> > > then pass NetCDF output of Grid-Stat to Point-Stat. And set up
> > > the Point-Stat configuration file to interpret that fractional
> > > coverage
> > field as a probability.
> > >
> > > In the long run, we should enhance Point-Stat to define
> > > neighborhood probabilities like this on the fly. But in the
short
> > > term, it sure would be nice to get the fractional field output
from Grid-Stat.
> > >
> > > But, as you've mentioned, you have already computed these
> > > probability values. What is the current format of the output?
We
> > > could probably write them to a NetCDF file that MET would
understand.
> > >
> > > John
> > >
> > > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via
RT
> > > < met_help at ucar.edu> wrote:
> > >
> > > >
> > > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
>
> > > >
> > > > Here is the list: for precip accum, total cloud cover, cloud
> > > > base hgt, and visibility, calculated the brier skill score
from
> > > > the model/ob pairs based on the forecasted neighborhoods. The
> > > > neighborhoods are
> > > defined below:
> > > >
> > > > For each ob, find nearest grid point, get all the grid points
in
> > > > the neighborhood surrounding this point, if they meet a
> > > > threshold for the variable set them to 1, else set to 0. Sum
> > > > the neighborhood grid points and divide by the number of
points.
> > > > This gives you a fcst "probability" to compare with the ob
> > > > point. The ob is set to 0 or 1 if the event occurred or
didn't
> > > > occur. So you end up with model ob pairs of the fcst
> > > > "probability" and the binary ob (typical ensemble
> > > verification).
> > > >
> > > > I have code that will do this. If I write out the results as
> > > > one of the MET output types, I was hoping I could then use
stat
> > > > analysis to compute Brier Skill Scores for these variables.
> > > >
> > > > The rest of the list: temperatures - calculate CRPSS from the
> > > > Neighborhood where each grid point in the neighborhood is
> > > > treated like an ensemble member. Wind speeds are converted to
> > > > the Beaufort Scale
> > > > (0-9) for the speed in the fcst neighborhoods and same for the
> > > > observation. So again, the fcst neighborhood is treated like
> > > > ensemble members and a RPSS is calculated.
> > > >
> > > > Supposedly, Barb Brown is familiar with HiRA according to
Marion
> > > > Mittermaier if you need more detail.
> > > >
> > > > Bob
> > > >
> > > > -----Original Message-----
> > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > <robert.craig.2 at us.af.mil>
> > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > >
> > > > Bob,
> > > >
> > > > The point-stat tool reads gridded forecast data and point
> observations.
> > > > So no, it cannot currently handle point observations.
> > > >
> > > > I'm not very familiar with the Marion's HiRA framework. What
> > > > are the final statistics you'd ultimately like to compute
using
> > > > these neighborhood probability forecasts and observation
pairs?
> > > > If we can pin down the statistics, we can figure out what
> > > > pathway through MET would make the most sense.
> > > >
> > > > Thanks,
> > > > John
> > > >
> > > >
> > > >
> > > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil via
> > > > RT < met_help at ucar.edu> wrote:
> > > >
> > > > >
> > > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > > Queue: met_help
> > > > > Subject: MET Input Formats
> > > > > Owner: Nobody
> > > > > Requestors: robert.craig.2 at us.af.mil
> > > > > Status: new
> > > > > Ticket <URL:
> > > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > > >
> > > > >
> > > > >
> > > > > John and company, As I mentioned in previous email, 557th
> > > > > really wants to make use of the HiRA framework proposed by
> > > > > Marion Mittermaier. I
> > > > plan to
> > > > > use MET as much as possible for this. The first step is to
> generate
> > > the
> > > > > model data neighborhood "probabilities" around each
> > > > > observation
> > point.
> > > > > I have written code to do this and the plan as to convert to
> > > > > NETCdf to feed in to MET. The question is can the FCST file
> > > > > for MET not be a gridded file when read into MET in NETCdf
format.
> > > > > If so, then MET could generate the model/ob pairs that end
up
> > > > > in MET point stat output. If MET can't handle point sourced
> > > > > FCST data, then I could generate the model/ob pairs myself
and
> > > > > then write out the data in a format that stat_analysis can
> > > > > read -
> maybe.
> > > > >
> > > > > Thanks
> > > > > Bob Craig
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: MET Input Formats
From: John Halley Gotway
Time: Tue Apr 26 14:30:25 2016
Bob,
My reply message failed to send because my attachment was too large.
Instead, I posted it here to our anonymous ftp site here:
ftp://ftp.rap.ucar.edu/incoming/irap/met_help/craig_data/sample_prob_mpr.txt
And here's my original message...
Bob,
Yes, I see what you're getting at.
Your code processes each point observation, looks at the gridded
forecast
values in the nearby neighborhood, applies a threshold, computes the
ratio
of grid points in the neighborhood meeting that threshold, and then
interprets that ratio as if it were a probabilistic forecast value.
So you basically have matched pairs of probabilistic forecasts and
deterministic observations. The only way to use them in MET would be
an
ASCII format the MPR output line from Point-Stat. Then you could pass
those pairs to the STAT-Analysis tool for the computation of
statistics.
By way of example, I've attached some similar MPR lines from running
Point-Stat to evaluate a probability of precip forecast. And here's a
stat-analysis job you can run on the attached file to compute
statistics:
stat_analysis -lookin ~/sample_prob_mpr.txt -job aggregate_stat \
-line_type MPR -out_line_type PSTD \
-out_fcst_thresh ge0.0,ge0.25,ge0.5,ge0.75,ge1.0 -out_obs_thresh
gt0 \
-out_stat sample_output_pstd.txt
This tells stat_analysis to read data from the -lookin file, only use
MPR
lines, aggregate them together and compute a PSTD output line, use the
-out_fcst_thresh list of probabilistic thresholds, use the single
-out_obs_thresh threshold to define the observed event, and then write
the
PSTD output to the file listed in -out_stat. FYI, you could also set
-out_line_type to PRC, PJC, and PCT for other output probabilistic
info.
If/when we implement this algorithm in MET, I can see two potential
ways of
doing it. One option would be doing it exactly as you describe.
Another
option would be to first pre-process the entire gridded forecast field
by
applying a single threshold and neighborhood size. The value at each
grid
point would be replaced by the fractional coverage of it's thresholded
neighborhood points. This is exactly what we do in Grid-Stat when
computing fractions skill score. Then just evaluate that gridded
field as
a probabilistic field in Point-Stat.
Both methods would yield very, very similar results, but there is a
subtle
difference in how the interpolation is performed. The latter method
would
fit very nicely in the existing logic/framework of the MET tools. The
former is doable but would require more custom code.
On Mon, Apr 25, 2016 at 6:51 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, the key question is the data file is not going to be gridded.
It
> will contain the model neighborhood values for each ob site. For
the HiRA
> framework, we compare the model neighborhood value calculated at
each obs
> site and compare to the ob value. I was hoping to use pointstat
for this
> but I was not sure of the NETCDF header parameters for a non-gridded
(or 1
> dimensional grid) data set.
>
> Thanks
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Friday, April 22, 2016 4:27 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> Unfortunately no, we don't have anything well-documented.
Eventually, we
> plan to switch from this very limited format to using CF-compliant
NetCDF.
>
> I'll describe the current format for you below. This is basically
the
> same format at the output of the pcp_combine tool (and all of the
other
> NetCDF output files MET writes). I've attached an example NetCDF
header
> from pcp_combine. Here are the relevant points:
>
> (1) Must define dimensions named "lat" and "lon" that match the grid
> dimensions.
> (2) You do *NOT* have to define corresponding variables for "lat"
and "lon"
> but you can if you'd like.
> (3) The data variable (e.g. "APCP_12" in the attached example) must
have
> dimensions (lat, lon).
> (4) Note that it's fine to define multiple variables and give them
> different names if you'd like.
> (5) Note that you can name the variables anything you'd like.
> (6) Specify variable attributes strings for: name, long_name, level,
and
> units.
> (7) Specify _FillValue attribute as a float value of -9999.
> (8) Specify timing info as strings: init_time_ut, valid_time_ut, and
> accum_time_sec.
> We store these as strings instead of integers to avoid overflow
> problems in NetCDF.
> FYI, you can specify strings for init_time, valid_time, and
> accum_time if you'd like, but MET does *NOT* actually read them.
> (9) In the global attributes, define: MET_version and Projection
> information (that's all the attributes after Projection and down to
ny).
>
> To see the NetCDF projection definition for your data, just run one
of
> your GRIB files through pcp_combine and look at the output. Use the
same
> projection settings in the NetCDF file you're creating.
>
> If you run into trouble, just let me know.
>
> Thanks,
> John
>
>
> On Fri, Apr 22, 2016 at 9:58 AM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > John, I am currently coding up the NETCDF file. I am not sure
what
> > global attributes I need to define for point data. Is there a
> > reference on what attributes MET expects in a NETCDF file for
different
> types of input data?
> >
> > Thanks
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 3:10 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > That sounds good to me. Once you have a sample NetCDF output
file,
> > feel free to send it to me. I can run it through Point-Stat and
send
> > you suggested configuration file to use.
> >
> > Thanks,
> > John
> >
> > On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > Okay, so I will write them out in NETCDF to look like output
from
> > > the pcp_combine tool. In point stat, this data will considered
the
> > > forecast data. Prep buffer will be the obs data. This would be
> > > great, since I can use MET for the stat calculations.
> > >
> > > Thanks
> > >
> > > Bob
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 2:36 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > I'd suggest writing them to NetCDF to make them look like the
NetCDF
> > > output from the pcp_combine tool. Yes, Point-Stat can read
NetCDF.
> > > We've really tried to abstract the reading of gridded data out
of
> > > the tools. As long as one MET tool can read a gridded data
file,
> > > all of them should be able to.
> > >
> > > And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> > > utility, which may be helpful.
> > >
> > > Thanks,
> > > John
> > >
> > > On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT
<
> > > met_help at ucar.edu> wrote:
> > >
> > > >
> > > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
>
> > > >
> > > > John, I am at the point where I am trying to decide what
format to
> > > > write the files out to. I could write the fcst neighborhoods
to
> > > > NetCDF if I can then feed them into MET for point stat. Or I
> > > > could go a step further and generate the model/ob pairs then
write
> > > > them out in a format that MET could use to calculate the
stats.
> > > > Or go a step further and calculate the statistics and then
write
> > > > them out in a format the MET stat analysis could read to
summarize
> > > > the data. My goal
> > > is to try to use MET in the greatest
> > > > extent possible here. Do you still have your hacked
grid_stat
> code?
> > > >
> > > > BOB
> > > >
> > > > -----Original Message-----
> > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > Sent: Wednesday, April 20, 2016 2:08 PM
> > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > <robert.craig.2 at us.af.mil>
> > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > >
> > > > Bob,
> > > >
> > > > The first step sounds identical to the computation of a
fractional
> > > > coverage fields in Grid-Stat. Grid-Stat computes fractional
> > > > coverage fields using a threshold and neighborhood size as the
> > > > first step in computing fractions skill score (and other
> > > > neighborhood
> > methods output).
> > > > Unfortunately, Grid-Stat does NOT currently write that
fractional
> > > > coverage field out to a gridded NetCDF file. Although, a few
> > > > years ago I hacked it to do so for one particular project.
> > > >
> > > > If Grid-Stat was writing those fields out, you could run the
field
> > > > through Grid-Stat to compute the gridded probability field and
> > > > then pass NetCDF output of Grid-Stat to Point-Stat. And set
up
> > > > the Point-Stat configuration file to interpret that fractional
> > > > coverage
> > > field as a probability.
> > > >
> > > > In the long run, we should enhance Point-Stat to define
> > > > neighborhood probabilities like this on the fly. But in the
short
> > > > term, it sure would be nice to get the fractional field output
from
> Grid-Stat.
> > > >
> > > > But, as you've mentioned, you have already computed these
> > > > probability values. What is the current format of the output?
We
> > > > could probably write them to a NetCDF file that MET would
understand.
> > > >
> > > > John
> > > >
> > > > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via
RT
> > > > < met_help at ucar.edu> wrote:
> > > >
> > > > >
> > > > > <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > > > >
> > > > > Here is the list: for precip accum, total cloud cover, cloud
> > > > > base hgt, and visibility, calculated the brier skill score
from
> > > > > the model/ob pairs based on the forecasted neighborhoods.
The
> > > > > neighborhoods are
> > > > defined below:
> > > > >
> > > > > For each ob, find nearest grid point, get all the grid
points in
> > > > > the neighborhood surrounding this point, if they meet a
> > > > > threshold for the variable set them to 1, else set to 0.
Sum
> > > > > the neighborhood grid points and divide by the number of
points.
> > > > > This gives you a fcst "probability" to compare with the ob
> > > > > point. The ob is set to 0 or 1 if the event occurred or
didn't
> > > > > occur. So you end up with model ob pairs of the fcst
> > > > > "probability" and the binary ob (typical ensemble
> > > > verification).
> > > > >
> > > > > I have code that will do this. If I write out the results
as
> > > > > one of the MET output types, I was hoping I could then use
stat
> > > > > analysis to compute Brier Skill Scores for these variables.
> > > > >
> > > > > The rest of the list: temperatures - calculate CRPSS from
the
> > > > > Neighborhood where each grid point in the neighborhood is
> > > > > treated like an ensemble member. Wind speeds are converted
to
> > > > > the Beaufort Scale
> > > > > (0-9) for the speed in the fcst neighborhoods and same for
the
> > > > > observation. So again, the fcst neighborhood is treated
like
> > > > > ensemble members and a RPSS is calculated.
> > > > >
> > > > > Supposedly, Barb Brown is familiar with HiRA according to
Marion
> > > > > Mittermaier if you need more detail.
> > > > >
> > > > > Bob
> > > > >
> > > > > -----Original Message-----
> > > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > > <robert.craig.2 at us.af.mil>
> > > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > > >
> > > > > Bob,
> > > > >
> > > > > The point-stat tool reads gridded forecast data and point
> > observations.
> > > > > So no, it cannot currently handle point observations.
> > > > >
> > > > > I'm not very familiar with the Marion's HiRA framework.
What
> > > > > are the final statistics you'd ultimately like to compute
using
> > > > > these neighborhood probability forecasts and observation
pairs?
> > > > > If we can pin down the statistics, we can figure out what
> > > > > pathway through MET would make the most sense.
> > > > >
> > > > > Thanks,
> > > > > John
> > > > >
> > > > >
> > > > >
> > > > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil
via
> > > > > RT < met_help at ucar.edu> wrote:
> > > > >
> > > > > >
> > > > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > > > Queue: met_help
> > > > > > Subject: MET Input Formats
> > > > > > Owner: Nobody
> > > > > > Requestors: robert.craig.2 at us.af.mil
> > > > > > Status: new
> > > > > > Ticket <URL:
> > > > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > > > >
> > > > > >
> > > > > >
> > > > > > John and company, As I mentioned in previous email, 557th
> > > > > > really wants to make use of the HiRA framework proposed by
> > > > > > Marion Mittermaier. I
> > > > > plan to
> > > > > > use MET as much as possible for this. The first step is
to
> > generate
> > > > the
> > > > > > model data neighborhood "probabilities" around each
> > > > > > observation
> > > point.
> > > > > > I have written code to do this and the plan as to convert
to
> > > > > > NETCdf to feed in to MET. The question is can the FCST
file
> > > > > > for MET not be a gridded file when read into MET in NETCdf
> format.
> > > > > > If so, then MET could generate the model/ob pairs that end
up
> > > > > > in MET point stat output. If MET can't handle point
sourced
> > > > > > FCST data, then I could generate the model/ob pairs myself
and
> > > > > > then write out the data in a format that stat_analysis
can
> > > > > > read -
> > maybe.
> > > > > >
> > > > > > Thanks
> > > > > > Bob Craig
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
Subject: RE: [rt.rap.ucar.edu #75996] MET Input Formats
From: robert.craig.2 at us.af.mil
Time: Wed Apr 27 08:40:37 2016
Okay, it looks like I have a path forward. Thanks for the help.
-----Original Message-----
From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
Sent: Tuesday, April 26, 2016 3:30 PM
To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
<robert.craig.2 at us.af.mil>
Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
Bob,
My reply message failed to send because my attachment was too large.
Instead, I posted it here to our anonymous ftp site here:
ftp://ftp.rap.ucar.edu/incoming/irap/met_help/craig_data/sample_prob_mpr.txt
And here's my original message...
Bob,
Yes, I see what you're getting at.
Your code processes each point observation, looks at the gridded
forecast values in the nearby neighborhood, applies a threshold,
computes the ratio of grid points in the neighborhood meeting that
threshold, and then interprets that ratio as if it were a
probabilistic forecast value.
So you basically have matched pairs of probabilistic forecasts and
deterministic observations. The only way to use them in MET would be
an ASCII format the MPR output line from Point-Stat. Then you could
pass those pairs to the STAT-Analysis tool for the computation of
statistics.
By way of example, I've attached some similar MPR lines from running
Point-Stat to evaluate a probability of precip forecast. And here's a
stat-analysis job you can run on the attached file to compute
statistics:
stat_analysis -lookin ~/sample_prob_mpr.txt -job aggregate_stat \
-line_type MPR -out_line_type PSTD \
-out_fcst_thresh ge0.0,ge0.25,ge0.5,ge0.75,ge1.0 -out_obs_thresh
gt0 \
-out_stat sample_output_pstd.txt
This tells stat_analysis to read data from the -lookin file, only use
MPR lines, aggregate them together and compute a PSTD output line, use
the -out_fcst_thresh list of probabilistic thresholds, use the single
-out_obs_thresh threshold to define the observed event, and then write
the PSTD output to the file listed in -out_stat. FYI, you could also
set -out_line_type to PRC, PJC, and PCT for other output probabilistic
info.
If/when we implement this algorithm in MET, I can see two potential
ways of doing it. One option would be doing it exactly as you
describe. Another option would be to first pre-process the entire
gridded forecast field by applying a single threshold and neighborhood
size. The value at each grid point would be replaced by the
fractional coverage of it's thresholded neighborhood points. This is
exactly what we do in Grid-Stat when computing fractions skill score.
Then just evaluate that gridded field as a probabilistic field in
Point-Stat.
Both methods would yield very, very similar results, but there is a
subtle difference in how the interpolation is performed. The latter
method would fit very nicely in the existing logic/framework of the
MET tools. The former is doable but would require more custom code.
On Mon, Apr 25, 2016 at 6:51 AM, robert.craig.2 at us.af.mil via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
>
> John, the key question is the data file is not going to be gridded.
> It will contain the model neighborhood values for each ob site. For
> the HiRA framework, we compare the model neighborhood value
calculated at each obs
> site and compare to the ob value. I was hoping to use pointstat
for this
> but I was not sure of the NETCDF header parameters for a non-gridded
> (or 1 dimensional grid) data set.
>
> Thanks
> Bob
>
> -----Original Message-----
> From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> Sent: Friday, April 22, 2016 4:27 PM
> To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> <robert.craig.2 at us.af.mil>
> Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
>
> Bob,
>
> Unfortunately no, we don't have anything well-documented.
Eventually,
> we plan to switch from this very limited format to using CF-
compliant NetCDF.
>
> I'll describe the current format for you below. This is basically
the
> same format at the output of the pcp_combine tool (and all of the
> other NetCDF output files MET writes). I've attached an example
> NetCDF header from pcp_combine. Here are the relevant points:
>
> (1) Must define dimensions named "lat" and "lon" that match the grid
> dimensions.
> (2) You do *NOT* have to define corresponding variables for "lat"
and "lon"
> but you can if you'd like.
> (3) The data variable (e.g. "APCP_12" in the attached example) must
> have dimensions (lat, lon).
> (4) Note that it's fine to define multiple variables and give them
> different names if you'd like.
> (5) Note that you can name the variables anything you'd like.
> (6) Specify variable attributes strings for: name, long_name, level,
> and units.
> (7) Specify _FillValue attribute as a float value of -9999.
> (8) Specify timing info as strings: init_time_ut, valid_time_ut, and
> accum_time_sec.
> We store these as strings instead of integers to avoid overflow
> problems in NetCDF.
> FYI, you can specify strings for init_time, valid_time, and
> accum_time if you'd like, but MET does *NOT* actually read them.
> (9) In the global attributes, define: MET_version and Projection
> information (that's all the attributes after Projection and down to
ny).
>
> To see the NetCDF projection definition for your data, just run one
of
> your GRIB files through pcp_combine and look at the output. Use the
> same projection settings in the NetCDF file you're creating.
>
> If you run into trouble, just let me know.
>
> Thanks,
> John
>
>
> On Fri, Apr 22, 2016 at 9:58 AM, robert.craig.2 at us.af.mil via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> >
> > John, I am currently coding up the NETCDF file. I am not sure
what
> > global attributes I need to define for point data. Is there a
> > reference on what attributes MET expects in a NETCDF file for
> > different
> types of input data?
> >
> > Thanks
> >
> > -----Original Message-----
> > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > Sent: Wednesday, April 20, 2016 3:10 PM
> > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > <robert.craig.2 at us.af.mil>
> > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> >
> > Bob,
> >
> > That sounds good to me. Once you have a sample NetCDF output
file,
> > feel free to send it to me. I can run it through Point-Stat and
> > send you suggested configuration file to use.
> >
> > Thanks,
> > John
> >
> > On Wed, Apr 20, 2016 at 1:45 PM, robert.craig.2 at us.af.mil via RT <
> > met_help at ucar.edu> wrote:
> >
> > >
> > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996 >
> > >
> > > Okay, so I will write them out in NETCDF to look like output
from
> > > the pcp_combine tool. In point stat, this data will considered
> > > the forecast data. Prep buffer will be the obs data. This
would
> > > be great, since I can use MET for the stat calculations.
> > >
> > > Thanks
> > >
> > > Bob
> > >
> > > -----Original Message-----
> > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > Sent: Wednesday, April 20, 2016 2:36 PM
> > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > <robert.craig.2 at us.af.mil>
> > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > >
> > > Bob,
> > >
> > > I'd suggest writing them to NetCDF to make them look like the
> > > NetCDF output from the pcp_combine tool. Yes, Point-Stat can
read NetCDF.
> > > We've really tried to abstract the reading of gridded data out
of
> > > the tools. As long as one MET tool can read a gridded data
file,
> > > all of them should be able to.
> > >
> > > And you'd be able to plot that NetCDF file using the
"plot_data_plane"
> > > utility, which may be helpful.
> > >
> > > Thanks,
> > > John
> > >
> > > On Wed, Apr 20, 2016 at 1:31 PM, robert.craig.2 at us.af.mil via RT
<
> > > met_help at ucar.edu> wrote:
> > >
> > > >
> > > > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
>
> > > >
> > > > John, I am at the point where I am trying to decide what
format
> > > > to write the files out to. I could write the fcst
neighborhoods
> > > > to NetCDF if I can then feed them into MET for point stat. Or
I
> > > > could go a step further and generate the model/ob pairs then
> > > > write them out in a format that MET could use to calculate the
stats.
> > > > Or go a step further and calculate the statistics and then
write
> > > > them out in a format the MET stat analysis could read to
> > > > summarize the data. My goal
> > > is to try to use MET in the greatest
> > > > extent possible here. Do you still have your hacked
grid_stat
> code?
> > > >
> > > > BOB
> > > >
> > > > -----Original Message-----
> > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > Sent: Wednesday, April 20, 2016 2:08 PM
> > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > <robert.craig.2 at us.af.mil>
> > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > >
> > > > Bob,
> > > >
> > > > The first step sounds identical to the computation of a
> > > > fractional coverage fields in Grid-Stat. Grid-Stat computes
> > > > fractional coverage fields using a threshold and neighborhood
> > > > size as the first step in computing fractions skill score (and
> > > > other neighborhood
> > methods output).
> > > > Unfortunately, Grid-Stat does NOT currently write that
> > > > fractional coverage field out to a gridded NetCDF file.
> > > > Although, a few years ago I hacked it to do so for one
particular project.
> > > >
> > > > If Grid-Stat was writing those fields out, you could run the
> > > > field through Grid-Stat to compute the gridded probability
field
> > > > and then pass NetCDF output of Grid-Stat to Point-Stat. And
set
> > > > up the Point-Stat configuration file to interpret that
> > > > fractional coverage
> > > field as a probability.
> > > >
> > > > In the long run, we should enhance Point-Stat to define
> > > > neighborhood probabilities like this on the fly. But in the
> > > > short term, it sure would be nice to get the fractional field
> > > > output from
> Grid-Stat.
> > > >
> > > > But, as you've mentioned, you have already computed these
> > > > probability values. What is the current format of the output?
> > > > We could probably write them to a NetCDF file that MET would
understand.
> > > >
> > > > John
> > > >
> > > > On Wed, Apr 20, 2016 at 12:51 PM, robert.craig.2 at us.af.mil via
> > > > RT < met_help at ucar.edu> wrote:
> > > >
> > > > >
> > > > > <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > > >
> > > > >
> > > > > Here is the list: for precip accum, total cloud cover, cloud
> > > > > base hgt, and visibility, calculated the brier skill score
> > > > > from the model/ob pairs based on the forecasted
neighborhoods.
> > > > > The neighborhoods are
> > > > defined below:
> > > > >
> > > > > For each ob, find nearest grid point, get all the grid
points
> > > > > in the neighborhood surrounding this point, if they meet a
> > > > > threshold for the variable set them to 1, else set to 0.
Sum
> > > > > the neighborhood grid points and divide by the number of
points.
> > > > > This gives you a fcst "probability" to compare with the ob
> > > > > point. The ob is set to 0 or 1 if the event occurred or
> > > > > didn't occur. So you end up with model ob pairs of the fcst
> > > > > "probability" and the binary ob (typical ensemble
> > > > verification).
> > > > >
> > > > > I have code that will do this. If I write out the results
as
> > > > > one of the MET output types, I was hoping I could then use
> > > > > stat analysis to compute Brier Skill Scores for these
variables.
> > > > >
> > > > > The rest of the list: temperatures - calculate CRPSS from
the
> > > > > Neighborhood where each grid point in the neighborhood is
> > > > > treated like an ensemble member. Wind speeds are converted
to
> > > > > the Beaufort Scale
> > > > > (0-9) for the speed in the fcst neighborhoods and same for
the
> > > > > observation. So again, the fcst neighborhood is treated
like
> > > > > ensemble members and a RPSS is calculated.
> > > > >
> > > > > Supposedly, Barb Brown is familiar with HiRA according to
> > > > > Marion Mittermaier if you need more detail.
> > > > >
> > > > > Bob
> > > > >
> > > > > -----Original Message-----
> > > > > From: John Halley Gotway via RT [mailto:met_help at ucar.edu]
> > > > > Sent: Wednesday, April 20, 2016 1:20 PM
> > > > > To: CRAIG, ROBERT J GS-12 USAF ACC 16 WS/WXN
> > > > > <robert.craig.2 at us.af.mil>
> > > > > Subject: Re: [rt.rap.ucar.edu #75996] MET Input Formats
> > > > >
> > > > > Bob,
> > > > >
> > > > > The point-stat tool reads gridded forecast data and point
> > observations.
> > > > > So no, it cannot currently handle point observations.
> > > > >
> > > > > I'm not very familiar with the Marion's HiRA framework.
What
> > > > > are the final statistics you'd ultimately like to compute
> > > > > using these neighborhood probability forecasts and
observation pairs?
> > > > > If we can pin down the statistics, we can figure out what
> > > > > pathway through MET would make the most sense.
> > > > >
> > > > > Thanks,
> > > > > John
> > > > >
> > > > >
> > > > >
> > > > > On Wed, Apr 20, 2016 at 10:27 AM, robert.craig.2 at us.af.mil
via
> > > > > RT < met_help at ucar.edu> wrote:
> > > > >
> > > > > >
> > > > > > Wed Apr 20 10:27:20 2016: Request 75996 was acted upon.
> > > > > > Transaction: Ticket created by robert.craig.2 at us.af.mil
> > > > > > Queue: met_help
> > > > > > Subject: MET Input Formats
> > > > > > Owner: Nobody
> > > > > > Requestors: robert.craig.2 at us.af.mil
> > > > > > Status: new
> > > > > > Ticket <URL:
> > > > > > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=75996
> > > > > > >
> > > > > >
> > > > > >
> > > > > > John and company, As I mentioned in previous email, 557th
> > > > > > really wants to make use of the HiRA framework proposed by
> > > > > > Marion Mittermaier. I
> > > > > plan to
> > > > > > use MET as much as possible for this. The first step is
to
> > generate
> > > > the
> > > > > > model data neighborhood "probabilities" around each
> > > > > > observation
> > > point.
> > > > > > I have written code to do this and the plan as to convert
to
> > > > > > NETCdf to feed in to MET. The question is can the FCST
file
> > > > > > for MET not be a gridded file when read into MET in NETCdf
> format.
> > > > > > If so, then MET could generate the model/ob pairs that end
> > > > > > up in MET point stat output. If MET can't handle point
> > > > > > sourced FCST data, then I could generate the model/ob
pairs
> > > > > > myself and then write out the data in a format that
> > > > > > stat_analysis can read -
> > maybe.
> > > > > >
> > > > > > Thanks
> > > > > > Bob Craig
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > > >
> > > >
> > > >
> > >
> > >
> > >
> > >
> >
> >
> >
> >
>
>
>
>
------------------------------------------------
More information about the Met_help
mailing list