[Met_help] [rt.rap.ucar.edu #94050] History for Neighborhood statistics with NMEP variables
John Halley Gotway via RT
met_help at ucar.edu
Wed Feb 5 15:46:43 MST 2020
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
Hi John,
I'm attempting to compute neighborhood statistics, specifically FSS, using
the new neighborhood maximum ensemble probability variables in the
EnsembleStat output. Unfortunately, I haven't been able to unlock the
config combination that results in the neighborhood line types being
computed and was hoping for some guidance at this point. The following is
my environment and GridStat executable on Jet (generated by
METplus wrappers):
*export MODEL="HRRRE"; export
FCST_VAR="APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1"; export
OBS_VAR="APCP_06"; export LEVEL="A06"; export OBTYPE="STAGEIV"; export
CONFIG_DIR="/home/amb-verif/ensemble/config/met"; export FCST_FIELD="{
name=\"APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1\"; level=\"A06\";
prob=TRUE; cat_thresh=[==0.1]; }"; export OBS_FIELD="{ name=\"APCP_06\";
level=\"A06\"; cat_thresh=[ >=12.7 ]; }"; export
INPUT_BASE="/home/amb-verif/ensemble/verif/precip/HRRRE/realtime"; export
MET_VALID_HHMM="0203"; export FCST_TIME="006"; export
MET_TMP_DIR="/home/amb-verif/ensemble/verif/precip/HRRRE/realtime/tmp";*
*/contrib/met/9.0_beta2/bin/grid_stat -v 2
/home/amb-verif/ensemble/verif/precip/HRRRE/realtime/20200203-18z/ensemble_stat/ensemble_stat_006_20200203_180000V_ens.nc
/home/amb-verif/ensemble/verif/precip/StageIV/realtime/20200203-18z/stageIV_06hr_precip.nc
/home/amb-verif/ensemble/config/met/GridStatConfig_APCP_ENS -outdir
/home/amb-verif/ensemble/verif/precip/HRRRE/realtime/20200203-18z/grid_stat*
In the MET config file, "GridStatConfig_APCP_ENS", I am running a Gaussian
smoother on the observations along with applying the same neighborhood
rules that were used in the creation of the NMEP variables in EnsembleStat:
*interp = {*
* field = OBS;*
* vld_thresh = 0.75;*
* shape = CIRCLE;*
* gaussian_dx = 3.0;*
* gaussian_radius = 25;*
* type = [ { method = GAUSSIAN; width = 1; } ];*
*}*
*nbrhd = {*
* field = OBS;*
* shape = CIRCLE;*
* width = [ 13 ];*
* cov_thresh = [ >=0.01 ];*
* vld_thresh = 0.75;*
*}*
What I was hoping this would do would be to calculate an obs neighborhood
fraction against the desired threshold that would allow proper comparison
to the NMEP variables being ingested. Unfortunately, it doesn't appear that
is what GridStat is doing, either when setting "prob=TRUE", which yields
empty neighborhood line types, or "prob_as_scalar=TRUE", which gives me an
apples-to-oranges result comparing the NMEP QPF probabilities to the actual
precip observed instead of the neighborhood frequency.
Sorry if this is unclear (let me know if I can clear anything up), but is
there a way that this can be done that I'm missing?
Thanks!
Jeff
--
-------------------------------------------------------------------------------
Jeff Hamilton
Jeffrey.A.Hamilton at noaa.gov
CIRES Research Associate phone: 303-497-7312
NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
325 Broadway, R/GSD1 DSRC Room 2B604
Boulder, CO 80305
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: John Halley Gotway
Time: Tue Feb 04 07:53:04 2020
Jeff,
Does what you’re trying to do actually make sense?
The output of the NMEP method in Ensemble-Stat are probability values
between 0 and 1. And you’re comparing that against 6-hour APCP from
Stage
IV.
You ultimately want to compute FSS, but do so using the raw NMEP
values
as-is versus the fractional coverage field computed over the
observations.
So in Grid-Stat, “process” the observations but not the forecast data.
The neighborhood methods are controlled by the nbrhd dictionary. Take
a
look in the users guide for a description of nbrhd.field. I think you
want
to set that to “OBS” instead of “BOTH”.
nbrhd = {
field = BOTH;
vld_thresh = 1.0;
shape = SQUARE;
width = [ 1 ];
cov_thresh = [ >=0.5 ];
}
Thanks,
John
On Tue, Feb 4, 2020 at 6:55 AM jeffrey.a.hamilton at noaa.gov via RT <
met_help at ucar.edu> wrote:
>
> Tue Feb 04 06:55:00 2020: Request 94050 was acted upon.
> Transaction: Ticket created by jeffrey.a.hamilton at noaa.gov
> Queue: met_help
> Subject: Neighborhood statistics with NMEP variables
> Owner: Nobody
> Requestors: jeffrey.a.hamilton at noaa.gov
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>
>
> Hi John,
>
> I'm attempting to compute neighborhood statistics, specifically FSS,
using
> the new neighborhood maximum ensemble probability variables in the
> EnsembleStat output. Unfortunately, I haven't been able to unlock
the
> config combination that results in the neighborhood line types being
> computed and was hoping for some guidance at this point. The
following is
> my environment and GridStat executable on Jet (generated by
> METplus wrappers):
>
> *export MODEL="HRRRE"; export
> FCST_VAR="APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1"; export
> OBS_VAR="APCP_06"; export LEVEL="A06"; export OBTYPE="STAGEIV";
export
> CONFIG_DIR="/home/amb-verif/ensemble/config/met"; export
FCST_FIELD="{
> name=\"APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1\";
level=\"A06\";
> prob=TRUE; cat_thresh=[==0.1]; }"; export OBS_FIELD="{
name=\"APCP_06\";
> level=\"A06\"; cat_thresh=[ >=12.7 ]; }"; export
> INPUT_BASE="/home/amb-verif/ensemble/verif/precip/HRRRE/realtime";
export
> MET_VALID_HHMM="0203"; export FCST_TIME="006"; export
> MET_TMP_DIR="/home/amb-
verif/ensemble/verif/precip/HRRRE/realtime/tmp";*
>
> */contrib/met/9.0_beta2/bin/grid_stat -v 2
>
> /home/amb-verif/ensemble/verif/precip/HRRRE/realtime/20200203-
18z/ensemble_stat/ensemble_stat_006_20200203_180000V_ens.nc
>
> /home/amb-verif/ensemble/verif/precip/StageIV/realtime/20200203-
18z/stageIV_06hr_
> precip.nc
> /home/amb-verif/ensemble/config/met/GridStatConfig_APCP_ENS -outdir
>
> /home/amb-verif/ensemble/verif/precip/HRRRE/realtime/20200203-
18z/grid_stat*
>
> In the MET config file, "GridStatConfig_APCP_ENS", I am running a
Gaussian
> smoother on the observations along with applying the same
neighborhood
> rules that were used in the creation of the NMEP variables in
EnsembleStat:
>
> *interp = {*
> * field = OBS;*
> * vld_thresh = 0.75;*
> * shape = CIRCLE;*
> * gaussian_dx = 3.0;*
> * gaussian_radius = 25;*
> * type = [ { method = GAUSSIAN; width = 1; } ];*
> *}*
> *nbrhd = {*
> * field = OBS;*
> * shape = CIRCLE;*
> * width = [ 13 ];*
> * cov_thresh = [ >=0.01 ];*
> * vld_thresh = 0.75;*
> *}*
>
> What I was hoping this would do would be to calculate an obs
neighborhood
> fraction against the desired threshold that would allow proper
comparison
> to the NMEP variables being ingested. Unfortunately, it doesn't
appear that
> is what GridStat is doing, either when setting "prob=TRUE", which
yields
> empty neighborhood line types, or "prob_as_scalar=TRUE", which gives
me an
> apples-to-oranges result comparing the NMEP QPF probabilities to the
actual
> precip observed instead of the neighborhood frequency.
>
> Sorry if this is unclear (let me know if I can clear anything up),
but is
> there a way that this can be done that I'm missing?
>
> Thanks!
>
> Jeff
>
>
> --
>
>
-------------------------------------------------------------------------------
> Jeff Hamilton
> Jeffrey.A.Hamilton at noaa.gov
> CIRES Research Associate phone: 303-497-7312
> NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
> 325 Broadway, R/GSD1 DSRC Room 2B604
> Boulder, CO 80305
>
>
------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: John Halley Gotway
Time: Tue Feb 04 08:00:32 2020
Ah, I see you are already using nbrhd.field = OBS in Grid-Stat.
So really you’re question is... are the interp settings applied first
prior
to applying the nbrhd settings? Or are they applied independently?
And
I’m not sure. I can check more closely if/when I get into work later
today.
John
On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
met_help at ucar.edu> wrote:
>
> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
> Transaction: Given to johnhg (John Halley Gotway) by RT_System
> Queue: met_help
> Subject: Neighborhood statistics with NMEP variables
> Owner: johnhg
> Requestors: jeffrey.a.hamilton at noaa.gov
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>
>
> This transaction appears to have no content
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #94050] Neighborhood statistics with NMEP variables
From: jeffrey.a.hamilton at noaa.gov
Time: Tue Feb 04 10:08:48 2020
Thanks, John. That will be very helpful to know. Looking at my earlier
email I should’ve simplified all of this down to the following
question: how do I create an observation frequency using neighborhood
methods that can be properly compared to the NMEP variables? That is
what I’m trying to do in GridStat in order to calculate FSS in this
case.
Thanks again,
Jeff
> On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT
<met_help at ucar.edu> wrote:
>
> Ah, I see you are already using nbrhd.field = OBS in Grid-Stat.
>
> So really you’re question is... are the interp settings applied
first prior
> to applying the nbrhd settings? Or are they applied independently?
And
> I’m not sure. I can check more closely if/when I get into work
later today.
>
> John
>
>> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
>> met_help at ucar.edu> wrote:
>>
>>
>> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
>> Transaction: Given to johnhg (John Halley Gotway) by RT_System
>> Queue: met_help
>> Subject: Neighborhood statistics with NMEP variables
>> Owner: johnhg
>> Requestors: jeffrey.a.hamilton at noaa.gov
>> Status: new
>> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>>
>>
>> This transaction appears to have no content
>>
>
------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: John Halley Gotway
Time: Tue Feb 04 11:42:23 2020
Jeff,
Perhaps you could do this as a multi-step process:
(1) Run Ensemble-Stat to compute NMEP (as you're already doing).
(2) Run Regrid-Data-Plane to pre-process the StageIV APCP_06 data by
"regridding" it to the same domain and applying whatever smoothing
method
should be used.
(3) Run Grid-Stat to compare NMEP to the output of Regrid-Data-Plane
and
compute FSS. And you'd keep nbrhd.field = OBS to only compute the
fractional coverage field for the observations.
Does that accomplish what you're trying to do? Full disclosure, I
haven't
thought through all the nitty-gritty details closely. But I'm hoping
that
Ensemble-Stat, Regrid-Data-Plane, and Grid-Stat can be strung together
to
accomplish this.
John
On Tue, Feb 4, 2020 at 10:09 AM jeffrey.a.hamilton at noaa.gov via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>
> Thanks, John. That will be very helpful to know. Looking at my
earlier
> email I should’ve simplified all of this down to the following
question:
> how do I create an observation frequency using neighborhood methods
that
> can be properly compared to the NMEP variables? That is what I’m
trying to
> do in GridStat in order to calculate FSS in this case.
>
> Thanks again,
>
> Jeff
>
> > On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT
<met_help at ucar.edu>
> wrote:
> >
> > Ah, I see you are already using nbrhd.field = OBS in Grid-Stat.
> >
> > So really you’re question is... are the interp settings applied
first
> prior
> > to applying the nbrhd settings? Or are they applied
independently? And
> > I’m not sure. I can check more closely if/when I get into work
later
> today.
> >
> > John
> >
> >> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
> >> met_help at ucar.edu> wrote:
> >>
> >>
> >> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
> >> Transaction: Given to johnhg (John Halley Gotway) by RT_System
> >> Queue: met_help
> >> Subject: Neighborhood statistics with NMEP variables
> >> Owner: johnhg
> >> Requestors: jeffrey.a.hamilton at noaa.gov
> >> Status: new
> >> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
> >>
> >>
> >> This transaction appears to have no content
> >>
> >
>
>
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #94050] Neighborhood statistics with NMEP variables
From: jeffrey.a.hamilton at noaa.gov
Time: Tue Feb 04 12:52:23 2020
Hi John,
In theory, that should do it and it is very similar to what I am doing
now. I currently run Regrid-Data-Plane to match the grids but leave
the Gaussian smoother to GridStat. I’ll give moving that step into
Regrid a shot. Unfortunately, Jet is down right now for maintenance,
but I should be able to test this tonight when it comes back.
Thanks again for your help and stay tuned!
Jeff
> On Feb 4, 2020, at 11:42 AM, John Halley Gotway via RT
<met_help at ucar.edu> wrote:
>
> Jeff,
>
> Perhaps you could do this as a multi-step process:
> (1) Run Ensemble-Stat to compute NMEP (as you're already doing).
> (2) Run Regrid-Data-Plane to pre-process the StageIV APCP_06 data by
> "regridding" it to the same domain and applying whatever smoothing
method
> should be used.
> (3) Run Grid-Stat to compare NMEP to the output of Regrid-Data-Plane
and
> compute FSS. And you'd keep nbrhd.field = OBS to only compute the
> fractional coverage field for the observations.
>
> Does that accomplish what you're trying to do? Full disclosure, I
haven't
> thought through all the nitty-gritty details closely. But I'm
hoping that
> Ensemble-Stat, Regrid-Data-Plane, and Grid-Stat can be strung
together to
> accomplish this.
>
> John
>
>> On Tue, Feb 4, 2020 at 10:09 AM jeffrey.a.hamilton at noaa.gov via RT
<
>> met_help at ucar.edu> wrote:
>>
>>
>> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>>
>> Thanks, John. That will be very helpful to know. Looking at my
earlier
>> email I should’ve simplified all of this down to the following
question:
>> how do I create an observation frequency using neighborhood methods
that
>> can be properly compared to the NMEP variables? That is what I’m
trying to
>> do in GridStat in order to calculate FSS in this case.
>>
>> Thanks again,
>>
>> Jeff
>>
>>> On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT
<met_help at ucar.edu>
>> wrote:
>>>
>>> Ah, I see you are already using nbrhd.field = OBS in Grid-Stat.
>>>
>>> So really you’re question is... are the interp settings applied
first
>> prior
>>> to applying the nbrhd settings? Or are they applied
independently? And
>>> I’m not sure. I can check more closely if/when I get into work
later
>> today.
>>>
>>> John
>>>
>>>> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
>>>> met_help at ucar.edu> wrote:
>>>>
>>>>
>>>> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
>>>> Transaction: Given to johnhg (John Halley Gotway) by RT_System
>>>> Queue: met_help
>>>> Subject: Neighborhood statistics with NMEP variables
>>>> Owner: johnhg
>>>> Requestors: jeffrey.a.hamilton at noaa.gov
>>>> Status: new
>>>> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>>>>
>>>>
>>>> This transaction appears to have no content
>>>>
>>>
>>
>>
>>
>
------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: jeffrey.a.hamilton at noaa.gov
Time: Wed Feb 05 12:32:58 2020
Hi John,
Unfortunately, moving the Gaussian smoother to Regrid_Data_Plane
didn't
have my desired result. The field it produces looks great, but I am
still
not getting any neighborhood statistic output in the *nbrcts*,
*nbrctc*,
and *nbrcnt *line types when running it through GridStat, which leads
me to
believe that I'm still not producing an observation frequency grid to
compare to the NMEP field. Here are my GridStat interp and nbrd
sections
now, knowing that regridding to the same 3km grid and the Gaussian
interpolation are done earlier:
*interp = {*
* field = BOTH;*
* vld_thresh = 0.75;*
* shape = SQUARE;*
* gaussian_dx = 3.0;*
* gaussian_radius = 25;*
* type = [ { method = NEAREST; width = 1; } ];*
*}*
*nbrhd = {*
* field = OBS;*
* shape = CIRCLE;*
* width = [ 13 ];*
* cov_thresh = [ >=0.01 ];*
* vld_thresh = 0.75;*
*}*
One thing I'm noticing is when GridStat runs, it gives me logging info
that
says it is computing Probabilistic Statistics, but doesn't say
anything
about Neighborhoods even though I have the output flags for all three
neighborhood line types set to true:
*DEBUG 2:
--------------------------------------------------------------------------------*
*DEBUG 2: *
*DEBUG 2: Processing APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
versus APCP_06A06, for smoothing method NEAREST(1), over region CONUS,
using 676103 matched pairs.*
*DEBUG 2: Computing Probabilistic Statistics.*
*DEBUG 2: Processing APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
versus APCP_06A06, for smoothing method NEAREST(1), over region EAST,
using
413185 matched pairs.*
*DEBUG 2: Computing Probabilistic Statistics.*
*DEBUG 2: *
*DEBUG 2:
--------------------------------------------------------------------------------*
Is there something else I need to do to enable neighborhood methods?
Thanks,
Jeff
On Tue, Feb 4, 2020 at 12:52 PM Jeffrey A Hamilton <
jeffrey.a.hamilton at noaa.gov> wrote:
> Hi John,
>
> In theory, that should do it and it is very similar to what I am
doing
> now. I currently run Regrid-Data-Plane to match the grids but leave
the
> Gaussian smoother to GridStat. I’ll give moving that step into
Regrid a
> shot. Unfortunately, Jet is down right now for maintenance, but I
should be
> able to test this tonight when it comes back.
>
> Thanks again for your help and stay tuned!
>
> Jeff
>
> > On Feb 4, 2020, at 11:42 AM, John Halley Gotway via RT <
> met_help at ucar.edu> wrote:
> >
> > Jeff,
> >
> > Perhaps you could do this as a multi-step process:
> > (1) Run Ensemble-Stat to compute NMEP (as you're already doing).
> > (2) Run Regrid-Data-Plane to pre-process the StageIV APCP_06 data
by
> > "regridding" it to the same domain and applying whatever smoothing
method
> > should be used.
> > (3) Run Grid-Stat to compare NMEP to the output of Regrid-Data-
Plane and
> > compute FSS. And you'd keep nbrhd.field = OBS to only compute the
> > fractional coverage field for the observations.
> >
> > Does that accomplish what you're trying to do? Full disclosure, I
> haven't
> > thought through all the nitty-gritty details closely. But I'm
hoping
> that
> > Ensemble-Stat, Regrid-Data-Plane, and Grid-Stat can be strung
together to
> > accomplish this.
> >
> > John
> >
> >> On Tue, Feb 4, 2020 at 10:09 AM jeffrey.a.hamilton at noaa.gov via
RT <
> >> met_help at ucar.edu> wrote:
> >>
> >>
> >> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
> >>
> >> Thanks, John. That will be very helpful to know. Looking at my
earlier
> >> email I should’ve simplified all of this down to the following
question:
> >> how do I create an observation frequency using neighborhood
methods that
> >> can be properly compared to the NMEP variables? That is what I’m
trying
> to
> >> do in GridStat in order to calculate FSS in this case.
> >>
> >> Thanks again,
> >>
> >> Jeff
> >>
> >>> On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT <
> met_help at ucar.edu>
> >> wrote:
> >>>
> >>> Ah, I see you are already using nbrhd.field = OBS in Grid-Stat.
> >>>
> >>> So really you’re question is... are the interp settings applied
first
> >> prior
> >>> to applying the nbrhd settings? Or are they applied
independently?
> And
> >>> I’m not sure. I can check more closely if/when I get into work
later
> >> today.
> >>>
> >>> John
> >>>
> >>>> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
> >>>> met_help at ucar.edu> wrote:
> >>>>
> >>>>
> >>>> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
> >>>> Transaction: Given to johnhg (John Halley Gotway) by RT_System
> >>>> Queue: met_help
> >>>> Subject: Neighborhood statistics with NMEP variables
> >>>> Owner: johnhg
> >>>> Requestors: jeffrey.a.hamilton at noaa.gov
> >>>> Status: new
> >>>> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050
> >
> >>>>
> >>>>
> >>>> This transaction appears to have no content
> >>>>
> >>>
> >>
> >>
> >>
> >
>
--
-------------------------------------------------------------------------------
Jeff Hamilton
Jeffrey.A.Hamilton at noaa.gov
CIRES Research Associate phone: 303-497-7312
NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
325 Broadway, R/GSD1 DSRC Room 2B604
Boulder, CO 80305
------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: John Halley Gotway
Time: Wed Feb 05 12:50:19 2020
Jeff,
Have you tried setting "prob_as_scalar = TRUE;"? The neighborhood
methods
are not applied when verifying probabilities... only scalar fields.
And if that doesn't work, please point me to your test directory on
jet and
I can take a closer look.
John
On Wed, Feb 5, 2020 at 12:33 PM jeffrey.a.hamilton at noaa.gov via RT <
met_help at ucar.edu> wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
>
> Hi John,
>
> Unfortunately, moving the Gaussian smoother to Regrid_Data_Plane
didn't
> have my desired result. The field it produces looks great, but I am
still
> not getting any neighborhood statistic output in the *nbrcts*,
*nbrctc*,
> and *nbrcnt *line types when running it through GridStat, which
leads me to
> believe that I'm still not producing an observation frequency grid
to
> compare to the NMEP field. Here are my GridStat interp and nbrd
sections
> now, knowing that regridding to the same 3km grid and the Gaussian
> interpolation are done earlier:
>
> *interp = {*
> * field = BOTH;*
> * vld_thresh = 0.75;*
> * shape = SQUARE;*
> * gaussian_dx = 3.0;*
> * gaussian_radius = 25;*
>
> * type = [ { method = NEAREST; width = 1; } ];*
> *}*
> *nbrhd = {*
> * field = OBS;*
> * shape = CIRCLE;*
> * width = [ 13 ];*
> * cov_thresh = [ >=0.01 ];*
> * vld_thresh = 0.75;*
> *}*
>
> One thing I'm noticing is when GridStat runs, it gives me logging
info that
> says it is computing Probabilistic Statistics, but doesn't say
anything
> about Neighborhoods even though I have the output flags for all
three
> neighborhood line types set to true:
>
> *DEBUG 2:
>
>
--------------------------------------------------------------------------------*
> *DEBUG 2: *
> *DEBUG 2: Processing
APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
> versus APCP_06A06, for smoothing method NEAREST(1), over region
CONUS,
> using 676103 matched pairs.*
> *DEBUG 2: Computing Probabilistic Statistics.*
> *DEBUG 2: Processing
APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
> versus APCP_06A06, for smoothing method NEAREST(1), over region
EAST, using
> 413185 matched pairs.*
> *DEBUG 2: Computing Probabilistic Statistics.*
> *DEBUG 2: *
> *DEBUG 2:
>
>
--------------------------------------------------------------------------------*
>
> Is there something else I need to do to enable neighborhood methods?
>
> Thanks,
>
> Jeff
>
> On Tue, Feb 4, 2020 at 12:52 PM Jeffrey A Hamilton <
> jeffrey.a.hamilton at noaa.gov> wrote:
>
> > Hi John,
> >
> > In theory, that should do it and it is very similar to what I am
doing
> > now. I currently run Regrid-Data-Plane to match the grids but
leave the
> > Gaussian smoother to GridStat. I’ll give moving that step into
Regrid a
> > shot. Unfortunately, Jet is down right now for maintenance, but I
should
> be
> > able to test this tonight when it comes back.
> >
> > Thanks again for your help and stay tuned!
> >
> > Jeff
> >
> > > On Feb 4, 2020, at 11:42 AM, John Halley Gotway via RT <
> > met_help at ucar.edu> wrote:
> > >
> > > Jeff,
> > >
> > > Perhaps you could do this as a multi-step process:
> > > (1) Run Ensemble-Stat to compute NMEP (as you're already doing).
> > > (2) Run Regrid-Data-Plane to pre-process the StageIV APCP_06
data by
> > > "regridding" it to the same domain and applying whatever
smoothing
> method
> > > should be used.
> > > (3) Run Grid-Stat to compare NMEP to the output of Regrid-Data-
Plane
> and
> > > compute FSS. And you'd keep nbrhd.field = OBS to only compute
the
> > > fractional coverage field for the observations.
> > >
> > > Does that accomplish what you're trying to do? Full disclosure,
I
> > haven't
> > > thought through all the nitty-gritty details closely. But I'm
hoping
> > that
> > > Ensemble-Stat, Regrid-Data-Plane, and Grid-Stat can be strung
together
> to
> > > accomplish this.
> > >
> > > John
> > >
> > >> On Tue, Feb 4, 2020 at 10:09 AM jeffrey.a.hamilton at noaa.gov via
RT <
> > >> met_help at ucar.edu> wrote:
> > >>
> > >>
> > >> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
> > >>
> > >> Thanks, John. That will be very helpful to know. Looking at my
earlier
> > >> email I should’ve simplified all of this down to the following
> question:
> > >> how do I create an observation frequency using neighborhood
methods
> that
> > >> can be properly compared to the NMEP variables? That is what
I’m
> trying
> > to
> > >> do in GridStat in order to calculate FSS in this case.
> > >>
> > >> Thanks again,
> > >>
> > >> Jeff
> > >>
> > >>> On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT <
> > met_help at ucar.edu>
> > >> wrote:
> > >>>
> > >>> Ah, I see you are already using nbrhd.field = OBS in Grid-
Stat.
> > >>>
> > >>> So really you’re question is... are the interp settings
applied first
> > >> prior
> > >>> to applying the nbrhd settings? Or are they applied
independently?
> > And
> > >>> I’m not sure. I can check more closely if/when I get into
work later
> > >> today.
> > >>>
> > >>> John
> > >>>
> > >>>> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT <
> > >>>> met_help at ucar.edu> wrote:
> > >>>>
> > >>>>
> > >>>> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
> > >>>> Transaction: Given to johnhg (John Halley Gotway) by
RT_System
> > >>>> Queue: met_help
> > >>>> Subject: Neighborhood statistics with NMEP variables
> > >>>> Owner: johnhg
> > >>>> Requestors: jeffrey.a.hamilton at noaa.gov
> > >>>> Status: new
> > >>>> Ticket <URL:
> https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050
> > >
> > >>>>
> > >>>>
> > >>>> This transaction appears to have no content
> > >>>>
> > >>>
> > >>
> > >>
> > >>
> > >
> >
>
>
> --
>
>
-------------------------------------------------------------------------------
> Jeff Hamilton
> Jeffrey.A.Hamilton at noaa.gov
> CIRES Research Associate phone: 303-497-7312
> NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
> 325 Broadway, R/GSD1 DSRC Room 2B604
> Boulder, CO 80305
>
>
------------------------------------------------
Subject: Neighborhood statistics with NMEP variables
From: jeffrey.a.hamilton at noaa.gov
Time: Wed Feb 05 14:27:59 2020
That did the trick! I'm getting very different values than I expected,
but
I'm betting it is due to my own misconfiguration. Will work through
them
and let you know if I discover anything.
One thing I'll note, the METplus regrid_data_plane wrapper does not
support
the Gaussian method (specifically adding gaussian_dx and
gaussian_radius as
argument options). I'm sure that is on the to-do list for future
versions,
but it would be nice to have at some point. In the meantime I have a
workaround in place.
Thanks again for all of your help,
Jeff
Thanks!
Jeff
On Wed, Feb 5, 2020 at 12:50 PM John Halley Gotway via RT
<met_help at ucar.edu>
wrote:
> Jeff,
>
> Have you tried setting "prob_as_scalar = TRUE;"? The neighborhood
methods
> are not applied when verifying probabilities... only scalar fields.
>
> And if that doesn't work, please point me to your test directory on
jet and
> I can take a closer look.
>
> John
>
> On Wed, Feb 5, 2020 at 12:33 PM jeffrey.a.hamilton at noaa.gov via RT <
> met_help at ucar.edu> wrote:
>
> >
> > <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050 >
> >
> > Hi John,
> >
> > Unfortunately, moving the Gaussian smoother to Regrid_Data_Plane
didn't
> > have my desired result. The field it produces looks great, but I
am still
> > not getting any neighborhood statistic output in the *nbrcts*,
*nbrctc*,
> > and *nbrcnt *line types when running it through GridStat, which
leads me
> to
> > believe that I'm still not producing an observation frequency grid
to
> > compare to the NMEP field. Here are my GridStat interp and nbrd
sections
> > now, knowing that regridding to the same 3km grid and the Gaussian
> > interpolation are done earlier:
> >
> > *interp = {*
> > * field = BOTH;*
> > * vld_thresh = 0.75;*
> > * shape = SQUARE;*
> > * gaussian_dx = 3.0;*
> > * gaussian_radius = 25;*
> >
> > * type = [ { method = NEAREST; width = 1; } ];*
> > *}*
> > *nbrhd = {*
> > * field = OBS;*
> > * shape = CIRCLE;*
> > * width = [ 13 ];*
> > * cov_thresh = [ >=0.01 ];*
> > * vld_thresh = 0.75;*
> > *}*
> >
> > One thing I'm noticing is when GridStat runs, it gives me logging
info
> that
> > says it is computing Probabilistic Statistics, but doesn't say
anything
> > about Neighborhoods even though I have the output flags for all
three
> > neighborhood line types set to true:
> >
> > *DEBUG 2:
> >
> >
>
--------------------------------------------------------------------------------*
> > *DEBUG 2: *
> > *DEBUG 2: Processing
APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
> > versus APCP_06A06, for smoothing method NEAREST(1), over region
CONUS,
> > using 676103 matched pairs.*
> > *DEBUG 2: Computing Probabilistic Statistics.*
> > *DEBUG 2: Processing
APCP_06_A06_ENS_NMEP_ge12.7_NBRHD169_GAUSSIAN1A06
> > versus APCP_06A06, for smoothing method NEAREST(1), over region
EAST,
> using
> > 413185 matched pairs.*
> > *DEBUG 2: Computing Probabilistic Statistics.*
> > *DEBUG 2: *
> > *DEBUG 2:
> >
> >
>
--------------------------------------------------------------------------------*
> >
> > Is there something else I need to do to enable neighborhood
methods?
> >
> > Thanks,
> >
> > Jeff
> >
> > On Tue, Feb 4, 2020 at 12:52 PM Jeffrey A Hamilton <
> > jeffrey.a.hamilton at noaa.gov> wrote:
> >
> > > Hi John,
> > >
> > > In theory, that should do it and it is very similar to what I am
doing
> > > now. I currently run Regrid-Data-Plane to match the grids but
leave the
> > > Gaussian smoother to GridStat. I’ll give moving that step into
Regrid a
> > > shot. Unfortunately, Jet is down right now for maintenance, but
I
> should
> > be
> > > able to test this tonight when it comes back.
> > >
> > > Thanks again for your help and stay tuned!
> > >
> > > Jeff
> > >
> > > > On Feb 4, 2020, at 11:42 AM, John Halley Gotway via RT <
> > > met_help at ucar.edu> wrote:
> > > >
> > > > Jeff,
> > > >
> > > > Perhaps you could do this as a multi-step process:
> > > > (1) Run Ensemble-Stat to compute NMEP (as you're already
doing).
> > > > (2) Run Regrid-Data-Plane to pre-process the StageIV APCP_06
data by
> > > > "regridding" it to the same domain and applying whatever
smoothing
> > method
> > > > should be used.
> > > > (3) Run Grid-Stat to compare NMEP to the output of Regrid-
Data-Plane
> > and
> > > > compute FSS. And you'd keep nbrhd.field = OBS to only compute
the
> > > > fractional coverage field for the observations.
> > > >
> > > > Does that accomplish what you're trying to do? Full
disclosure, I
> > > haven't
> > > > thought through all the nitty-gritty details closely. But I'm
hoping
> > > that
> > > > Ensemble-Stat, Regrid-Data-Plane, and Grid-Stat can be strung
> together
> > to
> > > > accomplish this.
> > > >
> > > > John
> > > >
> > > >> On Tue, Feb 4, 2020 at 10:09 AM jeffrey.a.hamilton at noaa.gov
via RT
> <
> > > >> met_help at ucar.edu> wrote:
> > > >>
> > > >>
> > > >> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050
>
> > > >>
> > > >> Thanks, John. That will be very helpful to know. Looking at
my
> earlier
> > > >> email I should’ve simplified all of this down to the
following
> > question:
> > > >> how do I create an observation frequency using neighborhood
methods
> > that
> > > >> can be properly compared to the NMEP variables? That is what
I’m
> > trying
> > > to
> > > >> do in GridStat in order to calculate FSS in this case.
> > > >>
> > > >> Thanks again,
> > > >>
> > > >> Jeff
> > > >>
> > > >>> On Feb 4, 2020, at 8:00 AM, John Halley Gotway via RT <
> > > met_help at ucar.edu>
> > > >> wrote:
> > > >>>
> > > >>> Ah, I see you are already using nbrhd.field = OBS in Grid-
Stat.
> > > >>>
> > > >>> So really you’re question is... are the interp settings
applied
> first
> > > >> prior
> > > >>> to applying the nbrhd settings? Or are they applied
independently?
> > > And
> > > >>> I’m not sure. I can check more closely if/when I get into
work
> later
> > > >> today.
> > > >>>
> > > >>> John
> > > >>>
> > > >>>> On Tue, Feb 4, 2020 at 7:53 AM The RT System itself via RT
<
> > > >>>> met_help at ucar.edu> wrote:
> > > >>>>
> > > >>>>
> > > >>>> Tue Feb 04 07:53:04 2020: Request 94050 was acted upon.
> > > >>>> Transaction: Given to johnhg (John Halley Gotway) by
RT_System
> > > >>>> Queue: met_help
> > > >>>> Subject: Neighborhood statistics with NMEP variables
> > > >>>> Owner: johnhg
> > > >>>> Requestors: jeffrey.a.hamilton at noaa.gov
> > > >>>> Status: new
> > > >>>> Ticket <URL:
> > https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=94050
> > > >
> > > >>>>
> > > >>>>
> > > >>>> This transaction appears to have no content
> > > >>>>
> > > >>>
> > > >>
> > > >>
> > > >>
> > > >
> > >
> >
> >
> > --
> >
> >
>
-------------------------------------------------------------------------------
> > Jeff Hamilton
> > Jeffrey.A.Hamilton at noaa.gov
> > CIRES Research Associate phone: 303-497-7312
> > NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
> > 325 Broadway, R/GSD1 DSRC Room 2B604
> > Boulder, CO 80305
> >
> >
>
>
--
-------------------------------------------------------------------------------
Jeff Hamilton
Jeffrey.A.Hamilton at noaa.gov
CIRES Research Associate phone: 303-497-7312
NOAA/OAR/ESRL/GSD/ADB fax: 303-497-4176
325 Broadway, R/GSD1 DSRC Room 2B604
Boulder, CO 80305
------------------------------------------------
More information about the Met_help
mailing list