[Wrfems] Comparing a 12 km run with 4.7 km run

Case, Jonathan (MSFC-VP61)[Other] jonathan.case-1 at nasa.gov
Thu Sep 17 07:17:06 MDT 2009

Hello Henry,

My response to you is based on the assumption that you ran with convective parameterization on the 12-km domain, and NO convective parameterization on the 4.7-km domain.  However, you didn't reveal any specific details on your physics choices, so I can't be sure.

If my assumption is correct, then this scenario is exactly what one might expect.  When convective parameterization is activated, it produces precipitation on the sub-grid scale without converting the precipitation over to microphysics species.  Since the reflectivity fields are derived by the post-processor based on the microphysics mixing ratios (i.e. rain water, snow, and graupel/hail), then you're going to see substantially less reflectivity when you use a convective parameterization scheme because only the grid-scale processes resulting in saturation and microphysics precipitation production will show up as forecast reflectivity.

Therefore, it's really not helpful to examine forecast reflectivity when convective parameterization is used on a coarser resolution grid.  You'd be better of looking at the accumulated precipitation between forecast output times.

I hope this helps to clarify the major differences.

Jonathan Case, ENSCO Inc.
NASA Short-term Prediction Research and Transition Center (aka SPoRT Center)
320 Sparkman Drive, Room 3062
Huntsville, AL 35805
Voice: 256.961.7504
Fax: 256.961.7788
Emails: Jonathan.Case-1 at nasa.gov / case.jonathan at ensco.com

"Whether the weather is cold, or whether the weather is hot, we'll weather
  the weather whether we like it or not!"

From: wrfems-bounces at comet.ucar.edu [mailto:wrfems-bounces at comet.ucar.edu] On Behalf Of Henry Steigerwaldt
Sent: Tuesday, September 15, 2009 3:07 PM
To: wrfems at comet.ucar.edu
Subject: [Wrfems] Comparing a 12 km run with 4.7 km run

To All:

We've seen some poor performances regarding simulated reflectivity output by our WRF-ARW (4.7 km) model with what really
happens. Also, I'd say many times the NSSL's WRF-ARW model (4 km) does a much better job at predicting precip intensities and
locations than our model does. We've tried to set up our local WRF similar to NSSL's regarding the various options available in the
setup routines (i.e. microphysics, radiation scheme, land surface model, etc.).

In correspondence with Dr. Bob, here's one thing he stated to improve our model output...
I suspect the answer is to use a larger domain, which is my recommendation  for most everything. The NSSL domain is
980x750 by 4km, which is probably much larger than yours and allows for small scale forcing to develop upstream of your
region and outside of your current computational domain. You can cheat a bit with a nested simulation going from 12 to 4km
with a large 12km domain. You can also start your run with a previous 6hour forecast to get additional time.

So, as an experiment, today we ran the model twice for the 12Z run to test out what Bob mentioned. The first run was the usual one with
a 4.7 km resolution and the small domain. The second run was then run with a much larger domain with a resolution of 12 km as Bob
recommended above. Attached is the result.

Focus on the differences within the area encompassed by the square, which is our usual domain size for the 4.7 km runs we produce.
Note that only part of the 12 km domain is visible in this screen capture. On the left is the 12 km version and on the right is the 4.7 km version.
Using this as an example, both show a 12-hour forecast valid at 00Z 16 SEP 2009 of the simulated Maximum Composite Reflectivity.

I was expecting to see at the very least a bit of improvement in forecast coverage of the precip in the bigger domain for reasons as Bob stated
above, than what was forecast in the much smaller domain we normally use. But I was startled to see such a difference between the two runs
in both areal coverage and intensities. They aren't even close! You would think the areas where precip is forecast would be generally similar,
and with the 12 km version showing less distinctness of the actual convective features of the echoes than the 4.7 km resolution version. For
other hours I looked at, it makes one think you were looking at runs for different time periods. They aren't similar at all.

What's going on with this? This can't possibly be something one can normally expect? I mean, the locations of the precip are so different,
and the same for the intensities, yet the runs are based on the same data and setup options for the most part. Only the size of the domains
and the resolutions are different as far as we can tell.

The purpose of this "experiment" was to determine whether perhaps using a larger domain for our model, or perhaps nesting a smaller domain
within a larger domain, would help improve our model precip forecasts.

Now I'm really confused! So what do we do?

Henry Steigerwaldt
SOO, WFO Nashville, TN
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.comet.ucar.edu/pipermail/wrfems/attachments/20090917/a1069479/attachment-0001.html>

More information about the Wrfems mailing list