Dear Jonathan:
I asked our remote sensing algorithm scientist to answer your question.
This is what he came up with. Note that in addition to the error estimate
value, which is the example I have provided, there is also a wind vector
quality indicator, which is a unitless value that can range from 0 to 100,
that is also discussed in his response immediately below.
The error estimate and quality indcator can be used to identify a subset of
high-quality DMW depending on the sensitivity of the user application
(e.g., assimilation into numerical weather prediction models).
Public reference:
http://www.star.nesdis.noaa.gov/goesr/product_winds_dmw.php
"Quality control of the retrieved DMWs is performed in two ways. The first
is through the application of target selection, feature tracking, and
height assignment error checks as described in the previous sections. The
second way involves the calculation of two quality indicators for each of
the DMWs using two different, but related, algorithms: the Quality
Indicator (QI) (Holmlund, 1998; Holmlund et al., 2001) and the Expected
Error (EE) (LeMarshall et al., 2004; Berger et al. 2008) "
"The statistically-based quality indicator (QI) developed at EUMETSAT
estimates the reliability of each Derived Motion Winds (DMW) based on
several quality control tests (Holmlund, 1998, Holmlund et. al 2001). These
tests not only analyze the consistency in space and time of each of the
intermediate DMW vector components, but also the height and temperature of
the tracers used in the vector determination, the symmetry of vector pairs
achieved from tracking tracers between consecutive images, differences with
surrounding vectors, and differences from a forecast field (optional). "
"The Expected Error (EE) algorithm, originally developed at the Australian
Bureau of Meteorology (LeMarshall et al, 2004) is an extension of the QI
algorithm described in the previous section. It is designed to express
quality in terms of a physical vector error metric (meters/second, m/s),
rather than a normalized score such as the QI."
"The outputted EE and QI quality indicators associated with each DMW
estimate can be used synergistically in order to optimize the quality and
geographic coverage of the final DMW dataset passed onto the user
community. The synergistic use of these quality indicators is designed to
take advantage of the strengths of each. The EE is superior at identifying
the quality of relatively slow DMWs, whereas the QI is better at
identifying the quality of relatively fast DMWs. A study conducted under
the GOES-R Risk Reduction (Berger et al. 2008) seeked to identify
thresholds for each parameter that could serve as a potential starting
point for users to use, if so desired, in any process they may have
established to select a subset of the highest quality DMWs. "
very respectfully,
randy
----------------------------------------
From: "Jonathan Gregory" <j.m.gregory at reading.ac.uk>
Sent: Monday, July 08, 2013 3:13 PM
To: cf-metadata at cgd.ucar.edu
Subject: Re: [CF-metadata] how to represent a non-standard error
Dear Randy
My question is perhaps the same as Jim's. The question is, how is this
number
interpreted, by those who use it? In what way does it quantify the error?
Best wishes
Jonathan
----- Forwarded message from "rhorne at excaliburlabs.com"
<rhorne at excaliburlabs.com> -----
> Date: Fri, 5 Jul 2013 13:43:25 -0400
> From: "rhorne at excaliburlabs.com" <rhorne at excaliburlabs.com>
> To: Jim Biard <jim.biard at noaa.gov>,
> "cf-metadata at cgd.ucar.edu List" <cf-metadata at cgd.ucar.edu>
> Subject: Re: [CF-metadata] how to represent a non-standard error
>
>
>
> Jim:
>
>
>
> An expected error value is individually calculated for EACH wind vector
in
> the product file based on the factors I identified a few emails back.
>
>
>
> This expected error allows the users of the GOES-R derived motion wind
> products to know what the error is for each wind vector in the product
> file. I think there are weather prediction models that make use of these
> error values. There may be other uses, but I don't know for sure.
>
>
>
>
>
> very respectfully,
>
>
>
> randy
>
> ----------------------------------------
> From: "Jim Biard" <jim.biard at noaa.gov>
> Sent: Friday, July 05, 2013 1:22 PM
> To: "cf-metadata at cgd.ucar.edu List" <cf-metadata at cgd.ucar.edu>
> Cc: "rhorne at excaliburlabs.com Horne" <rhorne at excaliburlabs.com>
> Subject: Re: [CF-metadata] how to represent a non-standard error
>
> Randy,
> Thanks. I guess I'm still wondering how you can validly assign a label
> such as "error" to a number that doesn't somehow tell you the likelihood
of
> the variable value being within some sort of interval. Without getting
too
> hung up on statistical formalisms, can you explain what use this number
is
> to you?
> Grace and peace,
> Jim
> Jim Biard
> Research Scholar
> Cooperative Institute for Climate and Satellites
> Remote Sensing and Applications Division
> National Climatic Data Center
> 151 Patton Ave, Asheville, NC 28801-5001
>
> jim.biard at noaa.gov
> 828-271-4900
> Follow us on Facebook!
> On Jul 5, 2013, at 1:13 PM, rhorne at excaliburlabs.com wrote:
>
> Jim:
>
> The magnitude of the Expected Error is a function of the calculated wind
> speed.
>
> In perusing the Expected Error algorithm documentation to compose the
last
> email, it appears the algorithm does not assume any type of an error
> distribution (normal or otherwise). The error estimate is absolute and is
> not associated with a confidence level.
>
> The use of this algorithm is based on the results it has achieved on
> predecessor weather satellite programs (empirical data has been used to
> determine its effectiveness.) I can provide you additional information on
> the Expected Error algorithm if you are interested,
>
> The point I am trying to make is that this is a specific error estimation
> approach that is unrelated to a sampling distribution. I would think
> there are others.
>
> very respectfully,
>
> randy
>
> From: "Jim Biard" <jim.biard at noaa.gov>
> Sent: Friday, July 05, 2013 11:22 AM
> To: "cf-metadata at cgd.ucar.edu List" <cf-metadata at cgd.ucar.edu>
> Cc: "rhorne at excaliburlabs.com Horne" <rhorne at excaliburlabs.com>
> Subject: Re: [CF-metadata] Fwd: how to represent a non-standard error
>
> Randy,
> Could you help me understand a touch more about this? You say it is an
> error that comes from a custom algorithm, but what defines what magnitude
> it has? How do you relate it to anything? Does it represent some sort of
> confidence interval?
> Grace and peace,
> Jim
> Jim Biard
> Research Scholar
> Cooperative Institute for Climate and Satellites
> Remote Sensing and Applications Division
> National Climatic Data Center
> 151 Patton Ave, Asheville, NC 28801-5001
>
> jim.biard at noaa.gov
> 828-271-4900
> <CicsLogoTiny.png>Follow us on Facebook!
> On Jul 5, 2013, at 9:28 AM, "rhorne at excaliburlabs.com"
> <rhorne at excaliburlabs.com> wrote:
>
> Dear Jonathan:
>
> In the case of the GOES-R derived motion winds product, the error
estimate
> (i.e. more formally referred to as Expected Error) is based on a custom
> algorithm.
>
> This expected error algorithm is specific to atmospheric wind vectors
> derived from satellte data. The overarching concept of the wind
algorithms
> generated from satellite data is doing pattern matching of phenomena
(like
> clouds) across multiple images of the same region separated by some
period
> of time
>
> The GOES-R incarnation of this Expected Error approach makes use of a set
> of error predictors including (1) NWP model data (wind shear, temperature
> gradient), (2) wind speed, direction, and consistency quality indicators
> output from the winds algorithm proper, and (3) a wavelength dependent
> constants (GOES-R generates sets of wind vectors from a visible and
several
> emissive bands)
>
> I also found an article on the web that discusses it:
>
>
https://www.eumetsat.int/cs/idcplg?IdcService=GET_FILE&dDocName=pdf_conf_p42
>
_s2_le_marshall&allowInterrupt=1&noSaveAs=1&RevisionSelectionMethod=LatestRe
> leased
>
> very respectfully,
>
> randy
>
> Dear all
>
> OK, I agree that if it's useful to compare them, then they should be
> described
> in a standardised way.
>
> Why is this *not* a standard error? I suppose that to be described as a
> standard error it should be a number you could regard as the standard
> deviation
> of the true value around the stated value. If it's not that, are there
> other
> ways you would use such a number?
>
> Best wishes
>
> Jonathan
>
> _______________________________________________
> CF-metadata mailing list
> CF-metadata at cgd.ucar.edu
> http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
>
> <CicsLogoTiny.png>
>
>
> _______________________________________________
> CF-metadata mailing list
> CF-metadata at cgd.ucar.edu
> http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
----- End forwarded message -----
_______________________________________________
CF-metadata mailing list
CF-metadata at cgd.ucar.edu
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <
http://mailman.cgd.ucar.edu/pipermail/cf-metadata/attachments/20130708/e8a24eeb/attachment-0001.html>
Received on Mon Jul 08 2013 - 15:02:27 BST
