[Met_help] intersection/union area question

John Halley Gotway johnhg at ucar.edu
Wed Mar 24 14:37:45 MDT 2010


Very good question.  I took a look at the output of METv1.1 versus METv2.0.  I just looked at the sample MODE output that's created by the test scripts that are distributed with MET.  I found that
there was a problem in the METv1.1 MODE output that's been corrected for METv2.0.

In METv1.1, it appears that there was a bug in how we were computing intersection area, union area, and symmetric difference.  Here's the files at which I'm looking:
   METv1.1 and METv2.0/out/mode/mode_APCP_12_SFC_vs_APCP_12_SFC_120000L_20050807_120000V_120000A_obj.txt

I've extracted some lines and columns of data from these files to demonstrate what we were doing wrong, and how it's been fixed:

F001      57   -9999             -9999      -9999          -9999
O002      4    -9999             -9999      -9999          -9999
F001_O002 -9999 0                42         42             0.00000

F001      57   NA                NA         NA             NA
O002      4    NA                NA         NA             NA
F001_O002 NA   0                 61         61             0.00000

In the example above, forecast object 1 (F001) has an area of 57 and observation object 2 (O002) has an area of 4.  In METv1.1, we were incorrectly listing the union area as 42 as the result of a bug.
 In METv2.0, we're now correctly listing the union area as 61.  Since these objects have 0 intersection, their union is just their sum.

As for how the intersection over area attribute is calculated, I admit I had to go look that up.  It's been a long time since I looked at it.  It's computed at the intersection area / minimum
(forecast area, observation area).  I see that on page 6-23 of the MET User's Guide we have that INCORRECTLY listed as intersection area divided by union area.  I'm make sure we correct that for the
next release.

If I recall correctly, I think we defined it in this way so that we could allow objects of very different sizes with a great deal of overlap to still match.  You can think of it this way - when
comparing two objects, compute the percentage of grid points matched for each object, and then take the larger of those percentages.

Hope that helps.  Feel free to write with more questions.

John Halley Gotway
johnhg at ucar.edu

Faye Barthold wrote:
> Hello,
> I am comparing MODE output from MET v1.0 and MET v2.0 (running on two 
> separate machines) in preparation to upgrade to MET v2.0.  All of the 
> output that I'm interested in is identical between the two versions with 
> the exception of the intersection area, union area, and ratio of the 
> intersection to union area.  In general, all three variables seem to 
> have higher values in the output from version 2.0.  For example:
> intersection area from MODE (v1.0) = 83
> union area from MODE (v1.0) = 326
> ratio from MODE (v1.0) = 0.722
> intersection area from MODE (v2.0) = 112
> union area from MODE (v2.0) = 387
> ratio from MODE (v2.0) = 0.974
> I've looked through the user guides and release notes for each version 
> and can't find anything that might indicate a change in how these values 
> are being calculated.  Is there a reason that these values would be 
> different between the two versions?  Since the areas seem to be larger 
> overall in v2.0, is that version somehow more sensitive and therefore 
> defining larger object areas?  Also, how is the ratio of the 
> intersection area to the union area being calculated?  I had assumed 
> that it was simply intersection area divided by union area, however the 
> values in the MODE output for the ratio don't make sense relative to the 
> individual intersection and union area values.  Using the v2.0 example 
> above, the ratio given by MODE is 0.974.  Calculating the ratio as 
> intersection area (112) divided by union area (387) gives a ratio of 
> only 0.289.  Am I missing something with this calculation?
> Thanks!
> Faye
> _______________________________________________
> Met_help mailing list
> Met_help at mailman.ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/met_help

More information about the Met_help mailing list