[Wrf-users] single grid point surface peculiarities in SNOW & TSLB out of real.exe (UNCLASSIFIED)

Dumais, Bob (Civ, ARL/CISD) robert.dumaisjr at us.army.mil
Mon Mar 15 16:51:19 MDT 2010


Classification: UNCLASSIFIED
Caveats: NONE

Sorry- should have forwarded to the  whole group.

                                          Bob
-------------------\
--------------------

Hi Mark,

Very nice to hear - and yes, I'd be interested in learning the steps of
how to do this. Thanks a bunch! It does appear the temps I am getting
currently are representative of the Indian Ocean, although I'm still not
too sure that is really where they come from via GFS. Honestly, I have
always been a bit confused when using NAM and GFS as to what the GRIB1
parameter 11/ level 1 (SKINTEMP) really is. It comes across as a value
that can be used for both skin temp over land and also SST if water. I
always figured it was just to be thought of as a "surface (land or
water)"  skin T prediction of the external model dependent on the
resolution of that model. If the external model is fine enough to
resolve some of the inland water bodies on your high res WRF grid, than
your initialization would be better. Otherwise, whatever land surface
T's the external model produced (assuming it was too course to resolve
your higher res water point(s)) would get used to initialize your inland
lake water T.  However, in my case the value of 296/297 K seems quite a
bit more than the GFS would have predicted for a 00Z land surface T over
that region on March 10. Maybe I'm wrong about that assumption, although
that is what led me to think the Indian Ocean for some reason (although
I am using SKINTEMP in my Vtable, and not SST). Thanks for whatever you
can provide to assist with the inland lake problem. Have a wonderful
evening-


                                               Bob


-----Original Message-----
From: Mark Stoelinga [mailto:mstoelinga at 3tier.com] 
Sent: Monday, March 15, 2010 4:26 PM
To: Dumais, Bob (Civ, ARL/CISD)
Cc: Min Zhu; Don Morton; David Ovens; Dr. Craig A. Mattocks; Sen Chiao;
Andre Pattantyus; Thomas Raab; Hoeth, Brian R. (JSC-WS8)[NOAA];
wrf-users at ucar.edu; Wang, Yansen (Civ, ARL/CISD)
Subject: Re: single grid point surface peculiarities in SNOW & TSLB out
of real.exe (UNCLASSIFIED)

Hi Bob,

I have a fix for this precise problem thanks to Michael Duda at NCAR.
WRF was producing "eternal lake effect storms" over lakes on the Tibetan
Plateau that were being assigned SST values representative of the Indian
Ocean...yikes!  But Michael's fix has eliminated the really bad LSTs and
all works fine now.

Unfortunately it is not a trivial thing...it involves transforming your
landuse data set to have a new "inland lakes" index (using code that
Michael Duda at NCAR has written) (or just obtaining the transformed
data from Michael or myself), making minor changes to GEOGRID.TBL and
namelist.wps, and running a new program after metgrid called
"tavgsfc.exe", whose output is then used by real.exe to define lake
temperatures as the diurnally averaged surface air temperature during
the time period of the metgrid analysis.

If you are using multiple resolutions of landuse, you would have to
transform all of them with Michael's code.  I've only used it on the 30s
landuse data set, and so now I use my transformed 30s landuse data with
lakes for all WRF grids of ~1km or coarser.

I can give you more details if you're interested.  I think I have an old
email somewhere that gives a step-by-step to someone else who asked
about it.

It would be really great if the NCAR WRF team could incorporate this fix
into the standard WPS and WRF systems in a manner that makes it much
more transparent to the user.  But I don't know if that is in their
queue or not.

Mark

--
Mark Stoelinga, PhD
Senior Scientist
3TIER Inc. <blockedhttp://www.3tier.com/> 3TIER Inc.
2001 6th Avenue, Suite 2100
Seattle, WA 98121
USA
+1 206.708.8588 direct
+1 206.325.1573 main
+1 206.708.8589 fax

www.3tier.com <blockedhttp://www.3tier.com/> 

This email message and all attachments transmitted with it are for the
sole use of the intended recipient(s) and may contain confidential
and/or legally privileged information. 

On Mar 15, 2010, at 1:42 PM, Dumais, Bob (Civ, ARL/CISD) wrote:


	Classification: UNCLASSIFIED
	Caveats: NONE
	
	FYI-
	
	        Refer to my recent emails to Dave Gill over at NCAR. It
looks
	to me like the problem of these occasional temperature
"bullseye"
	patterns arises due to the hi res land use interpolation
producing
	single point "islands" in the nest domain that are classified as
water.
	These isolated small inland water bodies are not well
initialized in
	terms of SST (or I guess SKINTEMP/TSK) values from the external
model
	(in the case of my Afghan 1 km nest example, the GFS). Is there
a way
	any of you know to improve the initial condition, without doing
too much
	effort? Maybe there is a parameter in GEOGRID.TBL or METGRID.TBL
that
	can allow for manual setting of SST/SKINTEMP/TSK for any nest
inland
	water bodies? Thanks-
	
	
	Bob
	
	-----Original Message-----
	From: Dumais, Bob (Civ, ARL/CISD) 
	Sent: Monday, March 15, 2010 2:07 PM
	To: 'gill at ucar.edu'
	Cc: 'Jim Dudhia'
	Subject: RE: single grid point surface peculiarities in SNOW &
TSLB out
	of real.exe (UNCLASSIFIED)
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	Dave,
	
	           Sorry to have bothered you so much on this topic! My
	conclusion after further looking at both the Afghanistan and the
	Washington state/Columbia River Basin hi res cases is this: 
	
	            (i) it is possible to get very small areas, even
singular
	grid point areas, of water classification within a much larger
	surrounding area of land (I suppose vice versa over an open
water area
	may also be possible). 
	
	            (ii) the default value for water temp (SST??) in
such an
	instance, such as from GFS 1-degree, may not at all be very
	representative. For example, the Afghanistan "water" grid point
where I
	have been finding the bullseye 2 magl temperature value pattern
is much
	warmer than it should be for this time of year (~ 296-297 deg K
	initialization). Over the CONUS, the higher res NAM 218 does a
decent
	job initializing SST/ Skin Temp for the larger inland lake
bodies such
	as Great Salt Lake.
	
	            (iii) I find that TSLB and SST fields at t=0 are the
same,
	but that TSLB becomes different thereafter. Does TSLB even
matter for a
	water point through a WRF simulation? Is it just garbage after
t=0 h?
	The T 2 magl does seem to stay fairly consistent with the
initial SST
	value throughout the recent Afghan simulation .
	
	            (iv) How best to modify in a cold start such initial
water
	temp values for points such as these, when it is obvious that
the
	initial value provided is out in left field? 
	
	
	
	Bob -----Original Message-----
	From: Dumais, Bob (Civ, ARL/CISD)
	Sent: Monday, March 15, 2010 1:23 PM
	To: 'gill at ucar.edu'
	Cc: 'Jim Dudhia'
	Subject: RE: single grid point surface peculiarities in SNOW &
TSLB out
	of real.exe (UNCLASSIFIED)
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	Hi again Dave,
	
	             I looked at this particular case in Afghanistan a
little
	more closely this morning, and it seems to actually be related
to the
	landuse / landmasking coming out of geogrid.tbl. These are high
	resolution nests (nest 1 is 9 km; nest 2 is 3 km; nest 3 is 1
km) and I
	used 30 arc sec terrain/land use option in geogrid for the
	interpolations of all my nests. I'm finding that for the point I
	identified on the 1 km nest (figures were attached in Friday's
email),
	the landmask and xland parameters also produce a bullseye due to
a
	"water" classification. This is basically a single "water" point
on the
	1 km nest out all alone amidst the surrounding land. When I
plotted
	"xland" using GRADS for the initial "0" output, it bullseyed to
a value
	of 1.79. I don't know if this is an artifact of GrADS, but
shouln't it
	have gone to 2? Also, plotting landmask produces a bulleseye
value of
	just under 0.3 (as opposed to 0 as I would expect). Finally,
plotting
	lu_index shows another bullseye pattern with 14 in the center
(ie;
	water). All this stuff leads to the behaviors in the SNOW & TSLB
field
	that I noted Friday. Any thoughts on this? Is it ok to leave as
a 1 km
	bullseye, even if it looks less than "pretty"? Thx-
	
	                                                            Bob 
	
	-----Original Message-----
	From: Dumais, Bob (Civ, ARL/CISD)
	Sent: Friday, March 12, 2010 5:34 PM
	To: 'gill at ucar.edu'
	Cc: 'Jim Dudhia'
	Subject: single grid point surface peculiarities in SNOW & TSLB
out of
	real.exe (UNCLASSIFIED)
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	Hi Dave,
	
	         I sent this email below off to Jimy a bit back, and he
	suggested forwarding to you since it seems a real.exe issue
(perhaps
	with snowcover adjustment if skintemp exceeds 0 deg C??). Enjoy
the
	weekend.
	
	
	Bob
	
	
	-----Original Message-----
	From: Dumais, Bob (Civ, ARL/CISD)
	Sent: Friday, March 12, 2010 5:07 PM
	To: 'Jim Dudhia'; wrfhelp
	Cc: 'Mark Stoelinga'; 'Sen Chiao'; 'Don Morton'; 'David Ovens';
'Andre
	Pattantyus'; 'Min Zhu'; Haines, Patrick (Civ, ARL/CISD); Knapp,
Dave
	(Civ, ARL/CISD); Wang, Yansen (Civ, ARL/CISD); Williamson, Chatt
(Civ,
	ARL/CISD); Passner, Jeff (Civ, ARL/CISD); 'Dr. Craig A.
Mattocks';
	'Thomas Raab'; 'Hoeth, Brian R. (JSC-WS8)[NOAA]'
	Subject: RE: [Wrf-users] Upper boundary cfl error (UNCLASSIFIED)
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	All,
	
	            I have captured some examples of the anomalous 2
magl T
	points (alluded to somewhere in our email trail below), which I
have
	come across in a few places where I have established high
resolution
	grids (such as the Columbia River Basin in WA & Afghanistan). I
	definately used GFS as ICs for my Afghanistan examples attached
here,
	but can't recall off the top of my head if it was NAM 218 or GFS
for the
	Washington state plots. 
	
	           The plots above show the evolution of three different
fields
	over time (SNOW, T2, & TSLB), the first set for a case over
Washington
	state and another for Afghanistan near Kabul. Both are "zoomed
in" areas
	of my 1 km domain to isolate the anomaly better. I also show
comparative
	temps at sigma level 1 in a few plots, to show that the anomaly
is
	basically isolated to the diagnostic 2magl T level and the
surface. I do
	realize now that I left out my t=0h TSLB plot for the Washington
case
	above.
	
	           What I've seen in these forecasts plotted is the
following
	behavior:
	
	                         (i) T2 starts out too warm at an
isolated grid
	point, and tend to remain that way
	
	                         (ii) TSLB starts out too warm at the
same
	isolated point (even though that plot is not attached, it does
so in the
	Washington case too), but then as the model evolves appears to
become a
	local cool spot. The Washington case started at 12z- the
Afghanistan
	case at 00 Z. 
	
	                          (iii) SNOW starts out with a local min
value
	at the same grid point.
	
	                          (iv) at the first sigma level and
above, this
	seems to have little if any impact on T field.
	
	          The WRF version used for the Washington run above was
	3.0.1.1, and 3.1.1 was used for Afghanistan. To me, this problem
seems
	rooted to the initial value of SNOW (and perhaps TSLB) that are
	generated by metgrid.exe in WPS 3.1. I looked at the METGRD.TBL
and I am
	using the default interpolation method of four pt + average_4pt.
I will
	try a different interpolation method next week to see if it has
an
	impact. Have any of you seen this before?
	
	
	Bob 
	
	
	
	
	
	
	
	-----Original Message-----
	From: Dumais, Bob (Civ, ARL/CISD)
	Sent: Tuesday, March 09, 2010 10:56 AM
	To: 'Jim Dudhia'
	Cc: Mark Stoelinga; Sen Chiao; Don Morton; David Ovens; Andre
	Pattantyus; Min Zhu; Haines, Patrick (Civ, ARL/CISD); Knapp,
Dave (Civ,
	ARL/CISD); Wang, Yansen (Civ, ARL/CISD); Williamson, Chatt (Civ,
	ARL/CISD); Passner, Jeff (Civ, ARL/CISD); Dr. Craig A. Mattocks;
Thomas
	Raab; Hoeth, Brian R. (JSC-WS8)[NOAA]
	Subject: RE: [Wrf-users] Upper boundary cfl error (UNCLASSIFIED)
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	Jimy,
	
	       I was thinking the same thing, and would be glad to!!
Very
	exciting I believe. I don't believe I'd be able to provide
anything like
	the rigid mathematical treatment you provided in your REPLY for
the
	various stability criteria, but I could attempt to provide a
little
	better information for users on how to apply epsmm and time step
for
	such a complex terrain region as British Columbia if applying a
fine
	horizontal resolution nesting scheme. For this, I would probably
stick
	initially to the 1 km grid spacing which gave me problems for
the past
	few months!  Thanks for all your help, as always. You are always
a good
	sport about supplying ideas and suggestions, even though I
realize you
	are probably swamped by your own research problems and concerns.
Have a
	great day!
	
	
	Bob
	
	
	-----Original Message-----
	From: Jim Dudhia [mailto:dudhia at ucar.edu]
	Sent: Tuesday, March 09, 2010 10:46 AM
	To: Dumais, Bob (Civ, ARL/CISD)
	Cc: Jim Dudhia; Mark Stoelinga; Sen Chiao; Don Morton; David
Ovens;
	Andre Pattantyus; Min Zhu; Haines, Patrick (Civ, ARL/CISD);
Knapp, Dave
	(Civ, ARL/CISD); Wang, Yansen (Civ, ARL/CISD); Williamson, Chatt
(Civ,
	ARL/CISD); Passner, Jeff (Civ, ARL/CISD); Dr. Craig A. Mattocks;
Thomas
	Raab; Hoeth, Brian R. (JSC-WS8)[NOAA]
	Subject: Re: [Wrf-users] Upper boundary cfl error (UNCLASSIFIED)
	
	Bob,
	   Very good to hear. I certainly encourage seeing if you can
get
	longer time steps by increasing epssm further. You should look
at
	whether other things in the solution are affected, but I think
probably
	not, since it is only sound waves we are damping heavily.
	   It might be nice to have a grid of various epssm and
time-steps to
	see what works and what doesn't, and give a clue how to
compromise
	between them. Could make for a nice workshop talk.
	Jimy
	
	On Mar 9, 2010, at 10:39 AM, Dumais, Bob (Civ, ARL/CISD) wrote:
	
	

		Classification: UNCLASSIFIED
		

		Caveats: NONE
		


		All,
		


		  Great news to report of my latest Whistler run!! Dr.
Dudhia's 
		

		suspicion seems confirmed, and his suggested solution
also seems to be
		


		spot on!! Last night, with the 90 level version of the
model on my 1 
		

		km domain, I was able to proceed with no issues at all!!
This with 
		

		nonlinear grid stagger per WRF DomainWizard, and a
lowest half-level 
		

		of about 24 magl. Recall that my outer mesh is 9 km (I
triple nest-
		

		9/3/1 km).
		


		               (i) set my course grid advective time
step to 6 s from
		


		my previous value of 9 s (then apply 3:1 ratio onto
other nests, so 
		

		the
		

		1 km nest has an advective time step of about 0.667 s) .
		


		               (ii) reset the value of the vertical
sound wave 
		

		damping coefficient (epssm) from 0.1 to 0.3
		


		  Now refer to my new plots above for the 1 km domain:
two of them at
		


		the 5h fcst mark valid 12-23-2010 17Z, and another
showing the w 
		

		vertical profile at the same point and time (50.0565 N;
-123.362 W, 30
		


		s at 12-23-2010 12:00:30 Z)that had the significant
low-level vertical
		


		instability in w yesterday. The differences are amazing!
The 
		

		combination of larger epsmm and smaller timestep
completely alleviated
		


		the problem.
		

		The 5 h forecast fields (I have just plotted wind
vectors here) look 
		

		very complex but reasonable. Jimy's earlier worked
showed that for 
		

		greater and greater model slopes, the solution had to be
either 
		

		progressively smaller advective time steps or higher
epssm (between 0 
		

		and 1). I kind of compromised here and did a little of
both, and the 
		

		results are very promising. I trust from Jimy's paper
that this 
		

		damping coeefficient has been shown to have limited to
no impacts on 
		

		the important flow/meteorological modes other than
acoustical.
		


		  My next step is to do the same model run again, except
with a 
		

		larger time step (9 s for 9 km grid). If this fails,
I'll keep that 
		

		time step and go to a higher epssm. Finally, if one or
the other 
		

		works, I'll try a final experiment with a time step of
27 s for 9 km 
		

		(a 3;1 ratio for grid space/time step). Many thanks need
to go out to 
		

		Dr. Dudhia for his suggestion over the past week-
someone should buy 
		

		him a beer this weekend!! We have plenty we can
experiment with 
		

		seemingly. Cheers-
		




		Bob
		


		-----Original Message-----
		

		From: Dumais, Bob (Civ, ARL/CISD)
		

		Sent: Monday, March 08, 2010 10:44 AM
		

		To: 'Jim Dudhia'; 'Mark Stoelinga'; 'Sen Chiao'; 'Don
Morton'; 'David 
		

		Ovens'; 'Andre Pattantyus'; 'Min Zhu'; Haines, Patrick
(Civ, ARL/ 
		

		CISD); Knapp, Dave (Civ, ARL/CISD); Wang, Yansen (Civ,
ARL/CISD); 
		

		Williamson, Chatt (Civ, ARL/CISD); Passner, Jeff (Civ,
ARL/CISD); 'Dr.
		

		Craig A.
		

		Mattocks'
		

		Subject: RE: Re: [Wrf-users] Upper boundary cfl error
(UNCLASSIFIED)
		


		Classification: UNCLASSIFIED
		

		Caveats: NONE
		


		Greetings!
		


		    For all of you who enjoyed last week's message trail
regarding 
		

		the single point "blow-up" on my Whistler, BC 1 km
domain- I have more
		

		:)
		


		    As noted in my earlier email from this morning, I
did move my 
		

		nest
		

		3 (1 km) domain center slightly to the east by 0.25 deg
longitude in 
		

		order to avoid that troublesome point in my initial
domain (~
		

		50.1286 N;
		

		-123.6265 W). The plots I emailed Thursday were from
that domain and 
		

		showing that particular grid point location.
		


		    I let the new domain with the shift run over the
weekend, and 
		

		when I looked at the results this morning I discovered
it blew up 
		

		again but at a new location!! At first glance it
appeared to be 
		

		possibly at the same I,J location on the new grid as in
the initial 
		

		domain. I have confirmed to myself this is not the case,
so we can put
		


		that concern to bed. This is back to purely a slope
discussion again I
		


		think.
		


		    The plots I have now show the blow up for the new
domain, at/near
		


		the grid point location (50.0565 N and -123.362 W). The
first set of 
		

		slides show the w, horizontal vector sfc wind,  and
terrain fields for
		


		this area/point. Notice the same exact behaviors as
noted for the 
		

		other location in my original configuration!! Since this
new grid 
		

		point location also happened to be contained in my
original nest 
		

		configuration, the last few plots attached here will
show some of 
		

		these same plots for the new location in the old
configuration. Notice
		


		the vertical profile of W in the last plot and how
similar it appears!
		

		Just
		

		not enough apparently to cause a CFL violation at this
point in the 
		

		original configuration. Interesting.
		


		     The only other comment I have to make about hi res
complex model
		


		topography and strange behaviors is this. In certain
places (such as 
		

		my Columbia River Basin runs in Washington at 1 km), I
do not see such
		


		blow ups but another phenomena. In or near the narrow
river basin 
		

		locations (which may be just a few grid points across),
I'll 
		

		occasionally find single points where there appear to be
"bullseyes"
		

		in the surface T/TD fields along with stronger wind
values. These 
		

		bullseyes appear almost immediately at the start of the
simulations, 
		

		and can cause differences of up to several deg C between
that point 
		

		and its surroundings. I do not know if there is some
kind of relation 
		

		to the things we are seeing with the slope, or if it
some kind of 
		

		Bernoulli flow kind of thing. I'll try to capture and
example for you 
		

		all later.
		


	
Bob
		


		PS: You can open all these gif attachments with MS
Picture Manager or 
		

		similar. Note that in my original configuration I sent
Thur, the blow 
		

		up really took off at about 1m 45s (even though my plots
may have 
		

		erroneously said 90s ). In my new configuration at the
new "blow up"
		

		location, the blow ups and CFLs appear to be even
earlier at 30s.  
		

		Plots
		

		1-6 attached show fields for this new point location.
Plots 7 and 8 
		

		show fields for this same location within the old
configuration at 1m 
		

		45s of that run. Notice it has some of that strange
low-level w 
		

		behavior but not enough to cause a blow up or CFL
violations!!
		

		Disregard the wording on my plot title in Figs 7-8 - it
should read 
		

		the opposite of what it does!
		



		-----Original Message-----
		

		From: Dumais, Bob (Civ, ARL/CISD)
		

		Sent: Thursday, March 04, 2010 5:22 PM
		

		To: 'Jim Dudhia'; 'Mark Stoelinga'; 'Sen Chiao'; 'Don
Morton'; David 
		

		Ovens; 'Andre Pattantyus'; 'Min Zhu'; Haines, Patrick
(Civ, ARL/CISD);
		


		Knapp, Dave (Civ, ARL/CISD); Wang, Yansen (Civ,
ARL/CISD); Williamson,
		


		Chatt (Civ, ARL/CISD); Passner, Jeff (Civ, ARL/CISD)
		

		Subject: Re: [Wrf-users] Upper boundary cfl error
(UNCLASSIFIED)
		


		Classification: UNCLASSIFIED
		

		Caveats: NONE
		


		All-
		


		            For anyone interested, here are a few GRADS
plots along 
		

		with my configuration namelists (using my triple nest
configuration) 
		

		for that problematic domain near Whistler, BC (where the
blow-ups 
		

		happen after a few minutes). I've tried several cases in
Feb, such as 
		

		Feb 23 12Z - Feb 24 12Z (using 06UTC Feb 23 GFS
datasets), and this 
		

		happens all the time in this configuration. I also have
a double nest 
		

		configuration (basically, nests 2 and 3 shown here
without outer nest,
		


		using
		

		NAM-218)
		

		where this happens also.  My GEOGRID.TBL shows I am
using just a 
		

		single pass smoothing in all my previous attempts.
Perhaps this is a 
		

		good domain for others to focus upon for investigating
this problem?
		

		Thanks,
		

		and all have a good evening!
		


		                                                 Bob
		

		Classification: UNCLASSIFIED
		

		Caveats: NONE
		



		Classification: UNCLASSIFIED
		

		Caveats: NONE
		



		Classification: UNCLASSIFIED
		

		Caveats: NONE
		



		<03fixw.gif><02fixw.gif><01fixw.gif>
		


	Classification: UNCLASSIFIED
	Caveats: NONE
	
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	
	
	Classification: UNCLASSIFIED
	Caveats: NONE
	


Classification: UNCLASSIFIED
Caveats: NONE




More information about the Wrf-users mailing list