[Met_help] MET Mode Question
John Halley Gotway
johnhg at ucar.edu
Mon Nov 9 12:48:58 MST 2009
Bob,
Let me answer your second question first. The "sum" of centroid_x and centroid_y is pretty meaningless. It's just the sum of all the values in those columns of data. However, "sum" would be
meaningful when you're looking at the distribution of object areas - it give you the total area for all the objects. For most attributes, it's meaningless, but for a handful of them it is very helpful.
For your first question, let's look at an example. Can you open up the following file generated by the MET test script:
METv2.0/out/mode/mode_APCP_12_SFC_vs_APCP_12_SFC_120000L_20050807_120000V_120000A.ps
On the bottom, right section of the first page you'll see columns labeled "Forecast" and "Observation". Look in those for the line "Simple/M/U". This gives you the counts of the matched and
unmatched simple objects. In the forecast field, there are 7 objects, 3 matched and 4 unmatched. In the observation field, there are 4 objects, 4 matched and 0 unmatched. The number of matched
simple objects differ because there is merging occurring in the forecast and observation fields. In the forecast field, the two red objects (#2 and #4) are merged together into a cluster object. In
the observation fields, the green objects (#2, #3, and #4) are merged into a cluster object. So that's why the number of matched/unmatched simple objects differ.
Now look at the link for "Cluster". You'll see that it's 2 in both the "Forecast" and "Observation" column. These values will always be the same. The number of forecast clusters by definition is
equal to the number of observation clusters. We don't even talk about unmatched clusters. Being matched or not only refers to simple objects. The clusters by definition are always matched.
Let me also point out that by changing the fcst_merge_flag, obs_merge_flag, and match_flag in the MODE config file, you can modify the logic that MODE uses to do matching/merging. These basically
allow you to modify how much merging of simple objects you'd like to allow.
Hope that helps.
Thanks,
John
Craig, Robert J Civ USAF AFWA 2 WXG/WEA wrote:
> John, thanks for the help. I am slowly getting the hang of this. A
> couple more questions I had concerning the matched vs unmatched fcst and
> obs:
>
> I assume that matched forecasts are all forecast objects that the
> software was able to match with an observation object and all matched
> observations are those observation objects matched to forecast objects.
> Why aren't the counts and areas for these the same?
>
> In the output file list below, what does sum refer to for centriod_x and
> y?
>
> Field N Min Max Mean StdDev P10 P25
> P50 P75 P90 Sum
> ---------- -- ---- ------- ------ ------ ----- -----
> ----- ------ ------ --------
> area 42 4.00 3502.00 360.93 660.14 15.70 50.00
> 99.00 386.00 823.00 15159.00
> centroid_x 42 1.50 151.50 78.41 54.08 7.10 17.71
> 86.80 132.09 143.98 3293.41
> centroid_y 42 1.50 118.50 64.64 43.43 8.43 18.51
> 69.73 110.07 117.49 2715.03
>
> Thanks
>
> Bob
>
>
> -----Original Message-----
> From: John Halley Gotway [mailto:johnhg at rap.ucar.edu]
> Sent: Friday, November 06, 2009 4:38 PM
> To: Craig, Robert J Civ USAF AFWA 2 WXG/WEA
> Cc: met_help at ucar.edu
> Subject: Re: [Met_help] MET Mode Question
>
> Bob,
>
> Sorry for the confusion. You're right that the MODE-Analysis tool
> options
> take some getting used to.
>
> I think the confusion is arising from the fact that some of the
> attributes
> in which your interested are defined for pairs of objects while other
> are
> defined only for single objects. You'll need to break those up into
> separate jobs.
>
> The following apply to single objects:
> CENTROID_X, CENTROID_Y, AREA
>
> And the following apply to pairs of objects:
> AREA_RATIO, INTERSECTION_AREA, UNION_AREA, PERCENTILE_INTENSITY_RATIO,
> INTEREST
>
> It sounds to me that you'd like to answer the following questions:
>
> (1) What do the distributions of AREA's and CENTROID_X,Y's for
> matched/unmatched simple forecast/observation objects look like?
>
> Run 4 jobs for each lead time, using these options:
>
> For matched fcst objects:
> -summary \
> -column AREA -column CENTROID_X -column CENTROID_Y \
> -simple -single -matched -fcst
>
> For unmatched fcst objects:
> -summary \
> -column AREA -column CENTROID_X -column CENTROID_Y \
> -simple -single -unmatched -fcst
>
> For matched obs objects:
> -summary \
> -column AREA -column CENTROID_X -column CENTROID_Y \
> -simple -single -matched -obs
>
> For unmatched obs objects:
> -summary \
> -column AREA -column CENTROID_X -column CENTROID_Y \
> -simple -single -unmatched -obs
>
> (2) How similar are the matched clusters of objects?
>
> Run 1 job for each lead time, using these options:
>
> For pairs of matched object clusters:
> -summary \
> -column AREA_RATIO -column INTERSECTION_AREA -column UNION_AREA \
> -column PERCENTILE_INTENSITY_RATIO -column INTEREST \
> -pair -cluster
>
> Hopefully that helps. Just let us know if you need any more assistance.
>
> Also, I noticed that you don't appear to be registered as a MET user in
> our database. When you get a chance, please fill out the registration
> form here:
> http://www.dtcenter.org/met/users/downloads/index.php
> Just enter you email address in the box and hit enter. That'll take you
> to the registration page.
>
> Thanks,
> John Halley Gotway
> johnhg at ucar.edu
>
>
>> We have been running MODE for our cloud forecasts over a period of
>> several weeks. Originally, we wanted to be able to summarize the
>> position errors of the cloud features. But our domain was too small
> to
>> be able to completely resolve cloud features so we are now running
> mode
>> on our larger domain forecast. However to make use of the data we
> have
>> already collected for a briefing, I was going to try to get a handle
> on
>> cloud area differences between the forecast and the observation.
> Also,
>> I would like to get the number of times the forecast missed the
> observed
>> cloud or cloud was forecast but didn't occur.
>>
>>
>>
>> Looking through the documentation, it just is not very clear on what
>> options I should set in mode analysis to get the data listed above.
> My
>> current command line is:
>>
>>
>>
>> /sphome/qcteam/METv2.0/bin/mode_analysis -lookin
>> /sphome/qcteam/METv2.0/out/mode/t06 -dump_row /sphome/qcteam/mode_dump
>> -summary -column AREA_RATIO -column INTERSECTION_AREA -column
> UNION_AREA
>> -column PERCENTILE_INTENSITY_RATIO -column CENTROID_X -column
> CENTROID_Y
>> -column AREA -column INTEREST -pair -cluster -out
>> /sphome/qcteam/dcf_obs_pair_cluster -fcst_lead 24
>>
>>
>>
>> I would like to summarize the data by forecast hour, so I would run it
>> with differenc -fcst_lead values. When I run this the area is 0. If
> I
>> run using -bycase, I get counts for fcst matched and obs matched but
> not
>> fcst unmatched and ob unmatched. I don't use the -matched/-unmatched
>> since the documentation implied you would get both if you didn't
> include
>> them.
>>
>>
>>
>> What options should I be running with?
>>
>>
>>
>> Thanks
>>
>>
>>
>> Bob Craig
>>
>> _______________________________________________
>> 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