[Met_help] [rt.rap.ucar.edu #60485] History for Clarification Reg the point rf
Tressa Fowler via RT
met_help at ucar.edu
Thu Aug 29 14:55:34 MDT 2013
----------------------------------------------------------------
Initial Request
----------------------------------------------------------------
Hi, I am verifyting WRF forecsat with the point rf. have following questions. 1. I read in the MET documentation that as far as the forecasts of the surface fields (say RF) are concerned, NO INTERPOLATION is done!!!! (Pg 4-1). They are MATCHED to Observations with the message type ADPSFC. There are 3-4 methods described in the PDf file how does the MET generates RF forecast at point location (nearest neighbour, min value, max value, least square etc). How do I know which method MET software has chosen/adopted for GENERATING the rainfall at the position of the rain-guage. 2. The method of least square is one kind of interpolation technique whereas it is mentioned that no interpolation is done for the surface fields. Pls explain!!!! 3. I want to know how does the ME, MAE etc and other scores change if the resolution of the model is changed from 30 to 15kms. I shall be waiting for your reply.
geeta
----------------------------------------------------------------
Complete Ticket History
----------------------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
From: John Halley Gotway
Time: Tue Mar 05 09:46:12 2013
Geeta,
Since you're comparing your gridded model output to point
observations, you should be using the Point-Stat tool. When verifying
surface fields against the ADPSFC (for surface) message type, no
vertical interpolation is performed. The horizontal interpolation is
performed in whatever way(s) you specify in the Point-Stat
configuration file. When you run Point-Stat, you specify one or more
horizontal interpolation method to use. Then, you should see output
statistics for each interpolation method you've chosen.
I'm not sure which version of MET you're using. If it's METv4.0,
here's the section of the configuration file where you select the
horizontal interpolation method(s). In the example below, I've
selected 3 methods:
//
// Interpolation methods
//
interp = {
vld_thresh = 1.0;
type = [
{
method = UW_MEAN;
width = 1;
},
{
method = DW_MEAN;
width = 2;
},
{
method = LS_FIT;
width = 2;
}
];
};
(1) The first is an unweighted-mean with a width of 1. That's just
the nearest-neighbor. No real interpolation is done, we just use the
forecast value at the grid point closest to the observation
location.
(2) The second is a distance-weighted mean with a width of 2. For
that, we use the forecast values at the 4 closest grid points and take
a weighted mean where the weights are 1 divided the
distance-squared.
(3) The third is a least-squares fit using the 4 closest grid boxes.
For more info on this, look at the file METv4.0/data/config/README.
So while no vertical interpolation is performed when verifying surface
fields using the ADPSFC message type, you have complete control over
which horizontal interpolation methods are applied.
Hope that helps.
Thanks,
John
On 03/03/2013 09:21 PM, Geeta Geeta via RT wrote:
>
> Sun Mar 03 21:21:05 2013: Request 60485 was acted upon.
> Transaction: Ticket created by geeta124 at hotmail.com
> Queue: met_help
> Subject: Clarification Reg the point rf
> Owner: Nobody
> Requestors: geeta124 at hotmail.com
> Status: new
> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=60485 >
>
>
> Hi, I am verifyting WRF forecsat with the point rf. have following
questions. 1. I read in the MET documentation that as far as the
forecasts of the surface fields (say RF) are concerned, NO
INTERPOLATION is done!!!! (Pg 4-1). They are MATCHED to Observations
with the message type ADPSFC. There are 3-4 methods described in the
PDf file how does the MET generates RF forecast at point location
(nearest neighbour, min value, max value, least square etc). How do I
know which method MET software has chosen/adopted for GENERATING the
rainfall at the position of the rain-guage. 2. The method of least
square is one kind of interpolation technique whereas it is mentioned
that no interpolation is done for the surface fields. Pls explain!!!!
3. I want to know how does the ME, MAE etc and other scores change if
the resolution of the model is changed from 30 to 15kms. I shall be
waiting for your reply.
> geeta
>
------------------------------------------------
Subject: Clarification Reg the point rf
From: Geeta Geeta
Time: Wed Mar 06 02:26:30 2013
Thanks John, I am using MEt 3.0. and I checked my configiration file,
I am using UW-MEAN. Further, I want to know which is the best method
for RF verification???. So what I should use like UW-MEAN, DISTANCE-
MEAN or least sq fit??? or any other. which is the best one amongst
all of these????
Also I want to know how does the ME, MAE etc and other scores change
if the resolution of the model is changed from 30 to 15kms. Pls give
me some insight into this. thanks
geeta
> Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point
rf
> From: met_help at ucar.edu
> To: geeta124 at hotmail.com
> CC: met_help-owner at mailman.ucar.edu; met_help at mailman.ucar.edu
> Date: Tue, 5 Mar 2013 09:46:13 -0700
>
> Geeta,
>
> Since you're comparing your gridded model output to point
observations, you should be using the Point-Stat tool. When verifying
surface fields against the ADPSFC (for surface) message type, no
> vertical interpolation is performed. The horizontal interpolation
is performed in whatever way(s) you specify in the Point-Stat
configuration file. When you run Point-Stat, you specify one or more
> horizontal interpolation method to use. Then, you should see output
statistics for each interpolation method you've chosen.
>
> I'm not sure which version of MET you're using. If it's METv4.0,
here's the section of the configuration file where you select the
horizontal interpolation method(s). In the example below, I've
> selected 3 methods:
>
> //
> // Interpolation methods
> //
> interp = {
> vld_thresh = 1.0;
>
> type = [
> {
> method = UW_MEAN;
> width = 1;
> },
> {
> method = DW_MEAN;
> width = 2;
> },
> {
> method = LS_FIT;
> width = 2;
> }
> ];
> };
>
> (1) The first is an unweighted-mean with a width of 1. That's just
the nearest-neighbor. No real interpolation is done, we just use the
forecast value at the grid point closest to the observation
> location.
> (2) The second is a distance-weighted mean with a width of 2. For
that, we use the forecast values at the 4 closest grid points and take
a weighted mean where the weights are 1 divided the
> distance-squared.
> (3) The third is a least-squares fit using the 4 closest grid boxes.
>
> For more info on this, look at the file METv4.0/data/config/README.
>
> So while no vertical interpolation is performed when verifying
surface fields using the ADPSFC message type, you have complete
control over which horizontal interpolation methods are applied.
>
> Hope that helps.
>
> Thanks,
> John
>
> On 03/03/2013 09:21 PM, Geeta Geeta via RT wrote:
> >
> > Sun Mar 03 21:21:05 2013: Request 60485 was acted upon.
> > Transaction: Ticket created by geeta124 at hotmail.com
> > Queue: met_help
> > Subject: Clarification Reg the point rf
> > Owner: Nobody
> > Requestors: geeta124 at hotmail.com
> > Status: new
> > Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=60485 >
> >
> >
> > Hi, I am verifyting WRF forecsat with the point rf. have following
questions. 1. I read in the MET documentation that as far as the
forecasts of the surface fields (say RF) are concerned, NO
INTERPOLATION is done!!!! (Pg 4-1). They are MATCHED to Observations
with the message type ADPSFC. There are 3-4 methods described in the
PDf file how does the MET generates RF forecast at point location
(nearest neighbour, min value, max value, least square etc). How do I
know which method MET software has chosen/adopted for GENERATING the
rainfall at the position of the rain-guage. 2. The method of least
square is one kind of interpolation technique whereas it is mentioned
that no interpolation is done for the surface fields. Pls explain!!!!
3. I want to know how does the ME, MAE etc and other scores change if
the resolution of the model is changed from 30 to 15kms. I shall be
waiting for your reply.
> > geeta
> >
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
From: John Halley Gotway
Time: Wed Mar 06 10:31:15 2013
Geeta,
I don't have an easy answer for you on this one. Often, users have a
preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
understand though that you're trying to compare the model performance
at 30km resolution to the model performance at 15km.
I'm going to refer this ticket to Tressa Fowler, the MET project lead,
who's a statistician. She may have some suggestions on this.
Thanks,
John
On 03/06/2013 02:26 AM, Geeta Geeta via RT wrote:
> Thanks John, I am using MEt 3.0. and I checked my configiration
file, I am using UW-MEAN. Further, I want to know which is the best
method for RF verification???. So what I should use like UW-MEAN,
DISTANCE-MEAN or least sq fit??? or any other. which is the best one
amongst all of these????
> Also I want to know how does the ME, MAE etc and other scores change
if the resolution of the model is changed from 30 to 15kms. Pls give
me some insight into this. thanks
> geeta
> > Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the
point rf
>> From: met_help at ucar.edu
>> To: geeta124 at hotmail.com
>> CC: met_help-owner at mailman.ucar.edu; met_help at mailman.ucar.edu
>> Date: Tue, 5 Mar 2013 09:46:13 -0700
>>
>> Geeta,
>>
>> Since you're comparing your gridded model output to point
observations, you should be using the Point-Stat tool. When verifying
surface fields against the ADPSFC (for surface) message type, no
>> vertical interpolation is performed. The horizontal interpolation
is performed in whatever way(s) you specify in the Point-Stat
configuration file. When you run Point-Stat, you specify one or more
>> horizontal interpolation method to use. Then, you should see
output statistics for each interpolation method you've chosen.
>>
>> I'm not sure which version of MET you're using. If it's METv4.0,
here's the section of the configuration file where you select the
horizontal interpolation method(s). In the example below, I've
>> selected 3 methods:
>>
>> //
>> // Interpolation methods
>> //
>> interp = {
>> vld_thresh = 1.0;
>>
>> type = [
>> {
>> method = UW_MEAN;
>> width = 1;
>> },
>> {
>> method = DW_MEAN;
>> width = 2;
>> },
>> {
>> method = LS_FIT;
>> width = 2;
>> }
>> ];
>> };
>>
>> (1) The first is an unweighted-mean with a width of 1. That's just
the nearest-neighbor. No real interpolation is done, we just use the
forecast value at the grid point closest to the observation
>> location.
>> (2) The second is a distance-weighted mean with a width of 2. For
that, we use the forecast values at the 4 closest grid points and take
a weighted mean where the weights are 1 divided the
>> distance-squared.
>> (3) The third is a least-squares fit using the 4 closest grid
boxes.
>>
>> For more info on this, look at the file METv4.0/data/config/README.
>>
>> So while no vertical interpolation is performed when verifying
surface fields using the ADPSFC message type, you have complete
control over which horizontal interpolation methods are applied.
>>
>> Hope that helps.
>>
>> Thanks,
>> John
>>
>> On 03/03/2013 09:21 PM, Geeta Geeta via RT wrote:
>>>
>>> Sun Mar 03 21:21:05 2013: Request 60485 was acted upon.
>>> Transaction: Ticket created by geeta124 at hotmail.com
>>> Queue: met_help
>>> Subject: Clarification Reg the point rf
>>> Owner: Nobody
>>> Requestors: geeta124 at hotmail.com
>>> Status: new
>>> Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=60485 >
>>>
>>>
>>> Hi, I am verifyting WRF forecsat with the point rf. have following
questions. 1. I read in the MET documentation that as far as the
forecasts of the surface fields (say RF) are concerned, NO
INTERPOLATION is done!!!! (Pg 4-1). They are MATCHED to Observations
with the message type ADPSFC. There are 3-4 methods described in the
PDf file how does the MET generates RF forecast at point location
(nearest neighbour, min value, max value, least square etc). How do I
know which method MET software has chosen/adopted for GENERATING the
rainfall at the position of the rain-guage. 2. The method of least
square is one kind of interpolation technique whereas it is mentioned
that no interpolation is done for the surface fields. Pls explain!!!!
3. I want to know how does the ME, MAE etc and other scores change if
the resolution of the model is changed from 30 to 15kms. I shall be
waiting for your reply.
>>> geeta
>>>
>>
>
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
From: John Halley Gotway
Time: Wed Mar 06 10:32:17 2013
Geeta,
I don't have an easy answer for you on this one. Often, users have a
preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
understand though that you're trying to compare the model performance
at 30km resolution to the model performance at 15km.
I'm going to refer this ticket to Tressa Fowler, the MET project lead,
who's a statistician. She may have some suggestions on this.
Thanks,
John
On 03/06/2013 10:31 AM, John Halley Gotway via RT wrote:
> Geeta,
>
> I don't have an easy answer for you on this one. Often, users have
a preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
> understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
>
> I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
>
> Thanks,
> John
------------------------------------------------
Subject: Clarification Reg the point rf
From: Geeta Geeta
Time: Wed Mar 06 21:29:46 2013
Thanks John, I am doing two things. 1. The model which is run
operationally (Officially) has a 9km resolution for which I am doing
the verification (RF) using point observations and using pointstat for
that. 2. I also have model outputs (my runs) at 15 and 30 km for
which I want to know how the ME and MAE and other scores like CSI etc
will change. In other words, Is is better to have runs at 30 or 15 km
keeping the ground truth the same ie same no of observations.
geeta
> Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point
rf
> From: met_help at ucar.edu
> To: geeta124 at hotmail.com
> CC: met_help-owner at mailman.ucar.edu; met_help at mailman.ucar.edu
> Date: Wed, 6 Mar 2013 10:32:17 -0700
>
> Geeta,
>
> I don't have an easy answer for you on this one. Often, users have
a preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
> understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
>
> I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
>
> Thanks,
> John
>
> On 03/06/2013 10:31 AM, John Halley Gotway via RT wrote:
> > Geeta,
> >
> > I don't have an easy answer for you on this one. Often, users
have a preference of one interpolation method over another for their
own reasons. But in general, one is not "better" than another. I
> > understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
> >
> > I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
> >
> > Thanks,
> > John
>
------------------------------------------------
Subject: Clarification Reg the point rf
From: Geeta Geeta
Time: Wed Mar 06 21:37:13 2013
Hi John, My point 1 was incomplete. I wished to know whether UW MEAN
or any other is used more widely???. What is the advantage of one over
the other. Eg. we have 5 methods in MET to get RF at the station
location of rain guage ie. Max, MIn, UW-Mean, DW-mean and Least Sq
fit. Why so many options were kept ???.Kindly give me some
clarification reg this.
geeta
From: geeta124 at hotmail.com
To: met_help at ucar.edu
Subject: RE: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
Date: Thu, 7 Mar 2013 09:59:44 +0530
Thanks John,
I am doing two things.
1. The model which is run operationally (Officially) has a 9km
resolution for which I am doing the verification (RF) using point
observations and using pointstat for that.
2. I also have model outputs (my runs) at 15 and 30 km for which I
want to know how the ME and MAE and other scores like CSI etc will
change.
In other words, Is is better to have runs at 30 or 15 km keeping the
ground truth the same ie same no of observations.
geeta
> Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
> From: met_help at ucar.edu
> To: geeta124 at hotmail.com
> CC: met_help-owner at mailman.ucar.edu; met_help at mailman.ucar.edu
> Date: Wed, 6 Mar 2013 10:32:17 -0700
>
> Geeta,
>
> I don't have an easy answer for you on this one. Often, users have
a preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
> understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
>
> I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
>
> Thanks,
> John
>
> On 03/06/2013 10:31 AM, John Halley Gotway via RT wrote:
> > Geeta,
> >
> > I don't have an easy answer for you on this one. Often, users
have a preference of one interpolation method over another for their
own reasons. But in general, one is not "better" than another. I
> > understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
> >
> > I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
> >
> > Thanks,
> > John
>
------------------------------------------------
Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
From: John Halley Gotway
Time: Thu Mar 07 13:03:43 2013
Geeta,
The following 3 types of interpolation are commonly used:
- nearest neighbor (UW_MEAN width 1)
- least-squares fit of the 4 closest points (LS_FIT width 2)
- distance-weighted mean of the 4 closest points (DW_MEAN width 2)
You're right that we do provide many interpolation options. We do
this to give the user flexibility in adapting the MET tools to a wide
range of verification tasks. While the 3 listed above are very
common, there are some cases when using a minimum or maximum would be
very appropriate.
Hope that helps.
John
On 03/06/2013 09:37 PM, Geeta Geeta via RT wrote:
>
> <URL: https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=60485 >
>
> Hi John, My point 1 was incomplete. I wished to know whether UW MEAN
or any other is used more widely???. What is the advantage of one over
the other. Eg. we have 5 methods in MET to get RF at the station
location of rain guage ie. Max, MIn, UW-Mean, DW-mean and Least Sq
fit. Why so many options were kept ???.Kindly give me some
clarification reg this.
>
> geeta
> From: geeta124 at hotmail.com
> To: met_help at ucar.edu
> Subject: RE: [rt.rap.ucar.edu #60485] Clarification Reg the point rf
> Date: Thu, 7 Mar 2013 09:59:44 +0530
>
>
>
>
> Thanks John,
> I am doing two things.
> 1. The model which is run operationally (Officially) has a 9km
resolution for which I am doing the verification (RF) using point
observations and using pointstat for that.
>
> 2. I also have model outputs (my runs) at 15 and 30 km for which I
want to know how the ME and MAE and other scores like CSI etc will
change.
> In other words, Is is better to have runs at 30 or 15 km keeping the
ground truth the same ie same no of observations.
>
>
> geeta
>
>> Subject: Re: [rt.rap.ucar.edu #60485] Clarification Reg the point
rf
>> From: met_help at ucar.edu
>> To: geeta124 at hotmail.com
>> CC: met_help-owner at mailman.ucar.edu; met_help at mailman.ucar.edu
>> Date: Wed, 6 Mar 2013 10:32:17 -0700
>>
>> Geeta,
>>
>> I don't have an easy answer for you on this one. Often, users have
a preference of one interpolation method over another for their own
reasons. But in general, one is not "better" than another. I
>> understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
>>
>> I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
>>
>> Thanks,
>> John
>>
>> On 03/06/2013 10:31 AM, John Halley Gotway via RT wrote:
>>> Geeta,
>>>
>>> I don't have an easy answer for you on this one. Often, users
have a preference of one interpolation method over another for their
own reasons. But in general, one is not "better" than another. I
>>> understand though that you're trying to compare the model
performance at 30km resolution to the model performance at 15km.
>>>
>>> I'm going to refer this ticket to Tressa Fowler, the MET project
lead, who's a statistician. She may have some suggestions on this.
>>>
>>> Thanks,
>>> John
>>
>
>
------------------------------------------------
Subject: Clarification Reg the point rf
From: Tressa Fowler
Time: Thu Mar 07 14:35:14 2013
Hi Geeta,
For precipitation, I would recommend using the nearest neighbor or
maximum. All other interpolation methods will smooth out the forecast,
a bad thing when many forecasts are 0. The smoothers make more sense
for continuous fields like temperature or pressure.
As for comparing 30 km to 15 km, I would identify a few event
thresholds of interest (like 2mm and 5mm) and use the fractions skill
score (FSS). This score can be computed for different neighborhood
sizes so that you can compare the skill on comparable scales.
Please see the MET documentation for further details about these
items.
Tressa
On Sun Mar 03 21:21:05 2013, geeta124 at hotmail.com wrote:
> Hi, I am verifyting WRF forecsat with the point rf. have following
> questions. 1. I read in the MET documentation that as far as the
> forecasts of the surface fields (say RF) are concerned, NO
> INTERPOLATION is done!!!! (Pg 4-1). They are MATCHED to
> Observations with the message type ADPSFC. There are 3-4 methods
> described in the PDf file how does the MET generates RF forecast
at
> point location (nearest neighbour, min value, max value, least
> square etc). How do I know which method MET software has
> chosen/adopted for GENERATING the rainfall at the position of the
> rain-guage. 2. The method of least square is one kind of
> interpolation technique whereas it is mentioned that no
> interpolation is done for the surface fields. Pls explain!!!! 3.
I
> want to know how does the ME, MAE etc and other scores change if
> the resolution of the model is changed from 30 to 15kms. I shall
> be waiting for your reply.
> geeta
------------------------------------------------
More information about the Met_help
mailing list