[Met_help] [rt.rap.ucar.edu #63076] History for question about percentiles of the errors.

[John Halley Gotway via RT]

----------------------------------------------------------------
  Initial Request
----------------------------------------------------------------

Dear:

I want to plot the box plots using the percentiles of the errors in NCl.
It need the extreme minimum, the extreme maximum and the median of each timeseries.
but the MET just output 10th, 25th, 50th, 75th, and 90th percentiles of the error
including bootstrap upper and lower confidence limits.
so my question :
1.how can get the the extreme minimum, the extreme maximum and the median of each timeseries.
2.the user's guide told the percentiles of the errors can provide more information than can be obtained
from the mean and standard deviations of the error, but it is different between the mean error and the standard deviations error in output(table 4-6). I am confused about which error the percentiles of the errors be obtained from in MET output.

Hoping your replay!

Best wishes!

Qiang



----------------------------------------------------------------
  Complete Ticket History
----------------------------------------------------------------

Subject: Re: [rt.rap.ucar.edu #63076] question about percentiles of the errors.
From: John Halley Gotway
Time: Mon Sep 23 11:54:53 2013

Qiang,

I apologize for the delay in responding to your question.

In order to create boxplots of the data from MET, you'll need more
than just the percentiles of the error.  You could use the 25th, 50th,
and 75th percentiles to plot the "box" part of the boxplot.
But to get the whiskers and any outliers, you'll need the individual
matched pair values.  While there are many variants on the boxplot out
there, the definition for a traditional "Tukey" boxplot is
that the whiskers extend to the largest data values that are not
"outliers".  An outlier is any point that falls outside of the median
+/- 1.5 * inter-quartile range.  To accurately define the
whiskers and outliers, you'll need the individual matched pair data.

Matched pair data is available from Point-Stat in the MPR (matched
pair) line type.  It's available from Grid-Stat in the gridded NetCDF
matched pair output file.  If you'd rather not deal with the
matched pair data, you could choose to create a line plot using the
percentiles of the error.  You could plot the median errors as a solid
line and the 25th and 75th percentiles as dashed lines
above/below the median.  Or you could plot the 10th and 90th instead.
That wouldn't show as much information as a boxplot, but would still
give an idea about the spread.

Regarding (2), I'm not sure exactly what your question is.  For each
matched pair, we have a forecast value, F, and an observation value,
O.  Then we compute an error as F - O.  We compute a mean
error (ME, column 53 of the CNT line) as the average of those error
values.  The standard deviation of the error (ESTDEV) is listed in
column 58 of the CNT line.  And the percentiles of the error
(E10, E25, E50, E75, and E90) are listed in columns 78 to 92.  The 2
mean and standard deviation values are the traditional ways of
summarizing the data.  But the 5 percentiles of the error give you a
bit more information about the distribution of values.  And the
boxplot is probably the most complete way to display the distribution
of errors.

Hope that helps clarify.

Thanks,
John


On 09/18/2013 03:07 PM, Li, Qiang via RT wrote:
>
> Wed Sep 18 15:07:22 2013: Request 63076 was acted upon.
> Transaction: Ticket created by liqiang at ou.edu
>         Queue: met_help
>       Subject: question about percentiles of the errors.
>         Owner: Nobody
>    Requestors: liqiang at ou.edu
>        Status: new
>   Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=63076 >
>
>
> Dear:
>
> I want to plot the box plots using the percentiles of the errors in
NCl.
> It need the extreme minimum, the extreme maximum and the median of
each timeseries.
> but the MET just output 10th, 25th, 50th, 75th, and 90th percentiles
of the error
> including bootstrap upper and lower confidence limits.
> so my question :
> 1.how can get the the extreme minimum, the extreme maximum and the
median of each timeseries.
> 2.the user's guide told the percentiles of the errors can provide
more information than can be obtained
> from the mean and standard deviations of the error, but it is
different between the mean error and the standard deviations error in
output(table 4-6). I am confused about which error the percentiles of
the errors be obtained from in MET output.
>
> Hoping your replay!
>
> Best wishes!
>
> Qiang
>

------------------------------------------------