[Met_help] [rt.rap.ucar.edu #87473] History for MODE Axis Angle Confidence

[John Halley Gotway via RT]

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

Hi there,

I read in the MODE v8.0 documentation that the interest computed for axis
angle difference is dependent on a confidence variable that is a function
of aspect ratio, but I couldn't find the exact relationship. I was
wondering if someone could tell me exactly how the confidence is calculated.

Thanks!
Jacob Radford


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

Subject: MODE Axis Angle Confidence
From: John Halley Gotway
Time: Mon Oct 22 15:16:23 2018

Hi Jacob,

I see you have a question about computing the confidence for the axis
angle
difference in MODE.

First, let me suggest that you try running MODE at verbosity level 5
to
dump out additional debug information.  For each fcst/obs object
comparison, MODE will write to standard out the following type of
information:
...
Angle Difference:
-----------------
   Value      = 57.266
   Interest   = 0.546
   Confidence = 0.446
   Weight     = 1.000
   Term       = 0.243
...
Total Interest = (sum of terms)/(sum of weights*confidence)
               = 8.405/9.295
               = 0.904

In this example, the angle difference was 57.266 degrees.  Here's the
interest function from the config file:

interest_function = {
...
   angle_diff = (
      (  0.0, 1.0 )
      ( 30.0, 1.0 )
      ( 90.0, 0.0 )
   );
...
}

Plugging 57.266 into this piecewise-linear function, we get an
interest
value = 0.546.
Next, the confidence is defined on the following lines of the file
engine.cc:

   2304    conf_obs    = aspect_ratio_conf(aspect_obs);
   2305    conf_fcst   = aspect_ratio_conf(aspect_fcst);
   2306    confidence  = sqrt(conf_obs*conf_fcst);

   3655 double aspect_ratio_conf(double t)
   3656
   3657 {
   3658
   3659 const double tm1   = t - 1.0;
   3660 const double ratio = (tm1*tm1)/(t*t + 1.0);
   3661
   3662 return( pow(ratio, 0.3) );
   3663
   3664 }

The more elongated an object is, the more confident we are in it's
axis
angle.  An aspect ratio (= width/length) of 1 means its circular.  The
closer to 0, the more elongated it is.

So we compute a confidence based on the aspect ratio of the forecast
and
observation objects.  The confidence in their comparison is the square
root
of their product.
In our working example, that's 0.446.  So the overall angle difference
term
= interest * confidence = 0.546 * 0.446 = 0.243.

And from the config file, the weight for this term is 1.0.

Does that make sense?

Thanks,
John

On Mon, Oct 22, 2018 at 2:55 PM Jacob Radford via RT
<met_help at ucar.edu>
wrote:

>
> Mon Oct 22 14:55:50 2018: Request 87473 was acted upon.
> Transaction: Ticket created by jtradfor at ncsu.edu
>        Queue: met_help
>      Subject: MODE Axis Angle Confidence
>        Owner: Nobody
>   Requestors: jtradfor at ncsu.edu
>       Status: new
>  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=87473 >
>
>
> Hi there,
>
> I read in the MODE v8.0 documentation that the interest computed for
axis
> angle difference is dependent on a confidence variable that is a
function
> of aspect ratio, but I couldn't find the exact relationship. I was
> wondering if someone could tell me exactly how the confidence is
> calculated.
>
> Thanks!
> Jacob Radford
>
>

------------------------------------------------
Subject: MODE Axis Angle Confidence
From: Jacob Radford
Time: Mon Oct 22 15:34:39 2018

Ahh, I see. I was missing this section where the aspect ratio
confidences
were defined:

   3655 double aspect_ratio_conf(double t)
   3656
   3657 {
   3658
   3659 const double tm1   = t - 1.0;
   3660 const double ratio = (tm1*tm1)/(t*t + 1.0);
   3661
   3662 return( pow(ratio, 0.3) );
   3663
   3664 }

Thanks for your help!

On Mon, Oct 22, 2018 at 5:16 PM John Halley Gotway via RT
<met_help at ucar.edu>
wrote:

> Hi Jacob,
>
> I see you have a question about computing the confidence for the
axis angle
> difference in MODE.
>
> First, let me suggest that you try running MODE at verbosity level 5
to
> dump out additional debug information.  For each fcst/obs object
> comparison, MODE will write to standard out the following type of
> information:
> ...
> Angle Difference:
> -----------------
>    Value      = 57.266
>    Interest   = 0.546
>    Confidence = 0.446
>    Weight     = 1.000
>    Term       = 0.243
> ...
> Total Interest = (sum of terms)/(sum of weights*confidence)
>                = 8.405/9.295
>                = 0.904
>
> In this example, the angle difference was 57.266 degrees.  Here's
the
> interest function from the config file:
>
> interest_function = {
> ...
>    angle_diff = (
>       (  0.0, 1.0 )
>       ( 30.0, 1.0 )
>       ( 90.0, 0.0 )
>    );
> ...
> }
>
> Plugging 57.266 into this piecewise-linear function, we get an
interest
> value = 0.546.
> Next, the confidence is defined on the following lines of the file
> engine.cc:
>
>    2304    conf_obs    = aspect_ratio_conf(aspect_obs);
>    2305    conf_fcst   = aspect_ratio_conf(aspect_fcst);
>    2306    confidence  = sqrt(conf_obs*conf_fcst);
>
>    3655 double aspect_ratio_conf(double t)
>    3656
>    3657 {
>    3658
>    3659 const double tm1   = t - 1.0;
>    3660 const double ratio = (tm1*tm1)/(t*t + 1.0);
>    3661
>    3662 return( pow(ratio, 0.3) );
>    3663
>    3664 }
>
> The more elongated an object is, the more confident we are in it's
axis
> angle.  An aspect ratio (= width/length) of 1 means its circular.
The
> closer to 0, the more elongated it is.
>
> So we compute a confidence based on the aspect ratio of the forecast
and
> observation objects.  The confidence in their comparison is the
square root
> of their product.
> In our working example, that's 0.446.  So the overall angle
difference term
> = interest * confidence = 0.546 * 0.446 = 0.243.
>
> And from the config file, the weight for this term is 1.0.
>
> Does that make sense?
>
> Thanks,
> John
>
> On Mon, Oct 22, 2018 at 2:55 PM Jacob Radford via RT
<met_help at ucar.edu>
> wrote:
>
> >
> > Mon Oct 22 14:55:50 2018: Request 87473 was acted upon.
> > Transaction: Ticket created by jtradfor at ncsu.edu
> >        Queue: met_help
> >      Subject: MODE Axis Angle Confidence
> >        Owner: Nobody
> >   Requestors: jtradfor at ncsu.edu
> >       Status: new
> >  Ticket <URL:
https://rt.rap.ucar.edu/rt/Ticket/Display.html?id=87473 >
> >
> >
> > Hi there,
> >
> > I read in the MODE v8.0 documentation that the interest computed
for axis
> > angle difference is dependent on a confidence variable that is a
function
> > of aspect ratio, but I couldn't find the exact relationship. I was
> > wondering if someone could tell me exactly how the confidence is
> > calculated.
> >
> > Thanks!
> > Jacob Radford
> >
> >
>
>

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