Introduce physics suites mechanism for ARW #124

mgduda · 2017-01-17T23:48:46Z

TYPE: enhancement

KEYWORDS: physics suites

SOURCE: internal

PURPOSE:
Introduce a mechanism for specifying physics suites, which automatically set physics options based on the suite selection.

DESCRIPTION OF CHANGES:
This PR introduces a mechanism for selecting suites of physics schemes via the namelist.input file, and it provides the initial definitions of two suites, tentatively named 'CONUS' and 'tropical'.

From a user perspective, the &physics namelist record contains a new option, physics_suite, which is a character string naming the physics suite to be used in a simulation. By default, namelist options for selecting individual physics schemes (specifically, mp_physics, cu_physics, ra_lw_physics, ra_sw_physics, bl_pbl_physics, sf_sfclay_physics, and sf_surface_physics) do not appear in the namelist. However, the selection of any of the aforementioned schemes may be overridden by specifying that namelist option explicitly.

At runtime, the model prints to the rsl files a summary of the physics schemes that will be used in the simulation:

*************************************
Configuring physics suite 'tropical'

         mp_physics:      6      6
         cu_physics:     16     16
      ra_lw_physics:      4      4
      ra_sw_physics:      4      4
     bl_pbl_physics:      1      1
  sf_sfclay_physics:     91     91
 sf_surface_physics:      2      2
*************************************

In the printout, if any options were overridden by the user, those options will be printed with a trailing asterisk.

The suite mechanism is implemented as a new routine, setup_physics_suite( ), that is called just after check_nml_consistency( ) and before set_physics_rconfigs( ). Suite definitions are hard-wired in this routine as CASE statements.

Several design decisions are worth mentioning:

Only a single physics suite is specified for a simulation; all nests are assumed to use the same physics options unless the user specifically overrides these options for a nested domain. For example, to override the cumulus scheme in domain 3, one might write:

 &physics
 physics_suite                       = 'tropical'
 cu_physics                          = -1,    -1,     0,
 ...

which results in the following printout to the rsl.error.0000 file:

*************************************
Configuring physics suite 'tropical'

         mp_physics:      6      6      6
         cu_physics:     16     16      0*
      ra_lw_physics:      4      4      4
      ra_sw_physics:      4      4      4
     bl_pbl_physics:      1      1      1
  sf_sfclay_physics:     91     91     91
 sf_surface_physics:      2      2      2

(* = option overrides suite setting)
*************************************

A value of -1 was chosen to indicate "the scheme following the choice of physics suite" for the namelist options mp_physics, cu_physics, ra_lw_physics, ra_sw_physics, etc., since a value of 0 is already used to indicate no parameterization.
Both the real.exe and wrf.exe programs may use physics suites, but at present no attempt to enable physics suites for WRF DA has been made.
The namelist.output file reflects the state of the namelist before processing of physics suites. This means that, for example, CU_PHYSICS= 11*-1 appears in the namelist.output file if the user has not explicitly overridden the selection of cumulus scheme for any domain.

LIST OF MODIFIED FILES:

main/module_wrf_top.F
main/real_em.F
Registry/Registry.EM_COMMON
share/module_check_a_mundo.F
test/em_real/namelist.input

TESTS CONDUCTED:
All WTF regression tests pass.

Short two-domain simulations with 'CONUS' and 'tropical' suites, with and without specific physics schemes overridden from suite default.

This option selects the physics suite, either 'continental_convection' or 'tropical_convection'

This new routine is responsible for setting physics options based on the physics suite selection made in the namelist.

The setting up of physics suites happens after the call to check_nml_consistency but before the call to set_physics_rconfigs.

The namelist variables mp_physics, ra_lw_physics, ra_sw_physics, sf_surface_physics, bl_pbl_physics, and cu_physics are set to -1 in the Registry.EM_COMMON file, indicating that these options should take on whatever value is dictated by the physics suite chosen by the user.

As part of a suite specification, we now include the surface layer scheme. The default value for the 'sf_sfclay_physics' namelist option is now -1, indicating that this option should default to the option from the selected suite.

…t.input All physics options that are selected by a suite should no longer appear in the default ARW namelist.input file.

When listing the physics schemes selected by a physics suite, the width of the field with the name of the option has been reduced.

jamiebresch · 2017-01-18T00:39:43Z

Why not have the default of physics_suite in Registry be 'none'?

davegill · 2017-01-18T00:49:27Z

@jamiebresch
Regarding physics suites in ARW, there is strong motivation in the WRF community that we are responding to. Specifically, there is interest to have a default suite of "works well together physics". After hearing this for the past couple of years at the WRF Workshop, the WRF and MPAS upper management basically required that this capability be part of this release. Michael's implementation follows the gist of the requirements that came out of several meetings. For new users, who will definitely be the primary beneficiaries of the suite capability, having "none" as the default sort of defeats the whole purpose.

mgduda · 2017-01-18T01:17:45Z

@mkavulich was just asking (in person) about whether we are changing the default behavior in WRF, and it's worth pointing out that using any existing namelist.input file should give the same results as would have been obtained before the introduction of physics suites, since existing namelist.input files explicitly specify (therefore, override) all physics options.

weiwangncar

Perhaps we should just use 'tropical' and 'continental' and remove 'convection' in the suite name.
We should also consider removing some of the non-critical physics namelists from our templates a later time.
Should we make recommendation for number of vertical levels too?

Rather than setting components of physics suites based on numbers that may change, we now use Registry-defined constants, e.g., 16 becomes NTIEDTKESCHEME for the selection of cu_physics.

mkavulich

Change names to "CONUS" and "tropical"
Make names case-insensitive

These two physics suites were previously known as 'continental_convection' and 'tropical_convection'.

Before the select-case block to handle physics suites, we first convert the name of the physics suite specified by the user to all lowercase.

mgduda · 2017-02-08T01:48:14Z

@mkavulich Both of your requested changes (name changes and case insensitivity) have been addressed.

jamiebresch · 2017-02-08T17:03:50Z

@mgduda the unlimited format string is Fortran 2008 standard.

mgduda · 2017-02-08T17:07:35Z

@jamiebresch Thanks for the info. Perhaps that explains why it works for various compilers besides PGI, and also why I couldn't find it in the Fortran 2003 book.

mgduda · 2017-02-08T17:08:55Z

@jamiebresch I'll update the commit message to mention Fortran 2003 and earlier, specifically.

The setup_physics_suite() routine previously used format strings like '(A21,*(I6,A1))' with the intention of repeating the (I6,A1) part some number of times determined at runtime. The use of '*' as a repetition count isn't valid in Fortran 90/95/2003, however, and was caught by the PGI compiler. This commit introduces code to build a format statement based on the runtime-determined repetition count, and to then use that format statement when printing out the schemes selected in a physics suite.

mgduda · 2017-02-08T21:38:19Z

@davegill and @dudhia you're both listed as reviewers for this PR. Can you add an approval if you have no objections to merging?

davegill

Approved

dudhia

As I mentioned to Michael, we need to think about the behavior when someone uses an old namelist which will be seen as 'CONUS' with a lot of overrides in the print-out.

mgduda · 2017-02-14T00:59:59Z

@dudhia @davegill @weiwangncar Jimy's concern about the use of existing namelists generating confusing printout in the rsl.error.0000 file is something that we may be able to address with relatively minor changes (described below). Perhaps we can have a quick vote: should we merge this PR as it is now to enable testing of suites to begin, or should we merge this PR after making changes to void confusing printout with old namelist files?

To avoid confusing printout with old namelist files, we could:

Set the default suite in the Registry.EM_COMMON file to none
In the suite setup routine, if the suite is none, print a simple message stating that we are not setting up a physics suite and return immediately

What do you all think?

dudhia · 2017-02-14T01:08:09Z

I prefer the default suite to be none. We can add a suite into the default em_real namelist.input. Having the default as none permits us to rename suites later without causing users problems or namelist-invisible changes to results. Jimy

…

On Mon, Feb 13, 2017 at 5:59 PM, Michael Duda ***@***.***> wrote: @dudhia <https://github.com/dudhia> @davegill <https://github.com/davegill> @weiwangncar <https://github.com/weiwangncar> Jimy's concern about the use of existing namelists generating confusing printout in the rsl.error.0000 file is something that we may be able to address with relatively minor changes (described below). Perhaps we can have a quick vote: should we merge this PR as it is now to enable testing of suites to begin, or should we merge this PR after making changes to void confusing printout with old namelist files? To avoid confusing printout with old namelist files, we could: 1. Set the default suite in the Registry.EM_COMMON file to none 2. In the suite setup routine, if the suite is none, print a simple message stating that we are not setting up a physics suite and return immediately What do you all think? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#124 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARGf_FYGLNbO9WQgVOW02SWMJam8bVoHks5rcPyPgaJpZM4LmTzn> .

weiwangncar · 2017-02-14T02:32:56Z

I'm ok with this proposal too. We can play with it once it's in and see what may need to be tweaked.

…

On Mon, Feb 13, 2017 at 6:08 PM, dudhia ***@***.***> wrote: I prefer the default suite to be none. We can add a suite into the default em_real namelist.input. Having the default as none permits us to rename suites later without causing users problems or namelist-invisible changes to results. Jimy On Mon, Feb 13, 2017 at 5:59 PM, Michael Duda ***@***.***> wrote: > @dudhia <https://github.com/dudhia> @davegill > <https://github.com/davegill> @weiwangncar > <https://github.com/weiwangncar> Jimy's concern about the use of existing > namelists generating confusing printout in the rsl.error.0000 file is > something that we may be able to address with relatively minor changes > (described below). Perhaps we can have a quick vote: should we merge this > PR as it is now to enable testing of suites to begin, or should we merge > this PR after making changes to void confusing printout with old namelist > files? > > To avoid confusing printout with old namelist files, we could: > > 1. Set the default suite in the Registry.EM_COMMON file to none > 2. In the suite setup routine, if the suite is none, print a simple > message stating that we are not setting up a physics suite and return > immediately > > What do you all think? > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#124 (comment)>, or mute > the thread > <https://github.com/notifications/unsubscribe-auth/ARGf_ FYGLNbO9WQgVOW02SWMJam8bVoHks5rcPyPgaJpZM4LmTzn> > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#124 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMHfkDU-93mPJqBMm3VVEY8dKurovlmnks5rcP57gaJpZM4LmTzn> .

This commit sets the default physics suite to 'none', and adds code to print a simple message and return early from the suite setup routine if the suite is 'none'. This should avoid any confusing output in the rsl files for anyone using existing namelist.input files with WRF v3.9.

mgduda · 2017-02-14T19:18:12Z

I've just pushed a commit that makes none the default physics suite in the Registry, and prints the following message if the physics suite is none:

*************************************
No physics suite selected.
Physics options will be used directly from the namelist.
*************************************

The default suite in the namelist at test/em_real/namelist.input is still CONUS.

If there are no objections, I'll rerun the regtest and merge this PR later today.

dudhia · 2017-02-14T19:24:09Z

What happens if someone mistypes a suite name. Does it abort with a message or default to something?

…

On Tue, Feb 14, 2017 at 12:18 PM, Michael Duda ***@***.***> wrote: I've just pushed a commit that makes none the default physics suite in the Registry, and prints the following message if the physics suite is none: ************************************* No physics suite selected. Physics options will be used directly from the namelist. ************************************* The default suite in the namelist at test/em_real/namelist.input is still CONUS. If there are no objections, I'll rerun the regtest and merge this PR later today. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#124 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARGf_HsEmvi54BTxh2nqOEmBPcTQ2IbRks5rcf31gaJpZM4LmTzn> .

mgduda · 2017-02-14T19:34:23Z

@dudhia yes, the code will stop with this fatal error:

-------------- FATAL CALLED ---------------
FATAL CALLED FROM FILE:  <stdin>  LINE:    1599
Unrecognized physics suite
-------------------------------------------

mgduda added 7 commits January 17, 2017 16:14

Add new namelist option 'physics_suite' to the &physics namelist record

9e13101

This option selects the physics suite, either 'continental_convection' or 'tropical_convection'

Add routine 'setup_physics_suite' to share/module_check_a_mundo.F

0a3844c

This new routine is responsible for setting physics options based on the physics suite selection made in the namelist.

Call setup_physics_suite for real.exe and wrf.exe

c44c950

The setting up of physics suites happens after the call to check_nml_consistency but before the call to set_physics_rconfigs.

Handle sf_sfclay_physics for suites

8666e59

As part of a suite specification, we now include the surface layer scheme. The default value for the 'sf_sfclay_physics' namelist option is now -1, indicating that this option should default to the option from the selected suite.

Remove physics options controlled by suites from test/em_real/namelis…

333d154

…t.input All physics options that are selected by a suite should no longer appear in the default ARW namelist.input file.

Reduce width of printout showing physics suite options

6d0df4b

When listing the physics schemes selected by a physics suite, the width of the field with the name of the option has been reduced.

mgduda requested review from davegill, weiwangncar and dudhia January 17, 2017 23:49

weiwangncar reviewed Jan 18, 2017

View reviewed changes

mgduda added 2 commits January 25, 2017 17:56

Change scheme numbers to Registry-defined constants

8f7709d

Rather than setting components of physics suites based on numbers that may change, we now use Registry-defined constants, e.g., 16 becomes NTIEDTKESCHEME for the selection of cu_physics.

Wrap lines longer than 132 characters in setup_physics_suite( )

36a4721

mkavulich suggested changes Feb 3, 2017

View reviewed changes

mgduda added 2 commits February 5, 2017 15:35

Change default physics suite names to 'CONUS' and 'tropical'

8a732fc

These two physics suites were previously known as 'continental_convection' and 'tropical_convection'.

Make physics suite names case insensitive

8cc6ce8

Before the select-case block to handle physics suites, we first convert the name of the physics suite specified by the user to all lowercase.

weiwangncar approved these changes Feb 7, 2017

View reviewed changes

mgduda force-pushed the physics_suites branch from b3098f3 to ba63baa Compare February 8, 2017 17:45

davegill approved these changes Feb 8, 2017

View reviewed changes

mkavulich approved these changes Feb 8, 2017

View reviewed changes

dudhia approved these changes Feb 10, 2017

View reviewed changes

mgduda merged commit 4962a48 into wrf-model:master Feb 15, 2017

mgduda deleted the physics_suites branch February 15, 2017 18:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Introduce physics suites mechanism for ARW #124

Introduce physics suites mechanism for ARW #124

mgduda commented Jan 17, 2017 •

edited

Loading

jamiebresch commented Jan 18, 2017

davegill commented Jan 18, 2017

mgduda commented Jan 18, 2017

weiwangncar left a comment

mkavulich left a comment •

edited by mgduda

Loading

mgduda commented Feb 8, 2017

jamiebresch commented Feb 8, 2017

mgduda commented Feb 8, 2017

mgduda commented Feb 8, 2017

mgduda commented Feb 8, 2017

davegill left a comment

dudhia left a comment

mgduda commented Feb 14, 2017

dudhia commented Feb 14, 2017 via email

weiwangncar commented Feb 14, 2017 via email

mgduda commented Feb 14, 2017

dudhia commented Feb 14, 2017 via email

mgduda commented Feb 14, 2017

Introduce physics suites mechanism for ARW #124

Introduce physics suites mechanism for ARW #124

Conversation

mgduda commented Jan 17, 2017 • edited Loading

jamiebresch commented Jan 18, 2017

davegill commented Jan 18, 2017

mgduda commented Jan 18, 2017

weiwangncar left a comment

Choose a reason for hiding this comment

mkavulich left a comment • edited by mgduda Loading

Choose a reason for hiding this comment

mgduda commented Feb 8, 2017

jamiebresch commented Feb 8, 2017

mgduda commented Feb 8, 2017

mgduda commented Feb 8, 2017

mgduda commented Feb 8, 2017

davegill left a comment

Choose a reason for hiding this comment

dudhia left a comment

Choose a reason for hiding this comment

mgduda commented Feb 14, 2017

dudhia commented Feb 14, 2017 via email

weiwangncar commented Feb 14, 2017 via email

mgduda commented Feb 14, 2017

dudhia commented Feb 14, 2017 via email

mgduda commented Feb 14, 2017

mgduda commented Jan 17, 2017 •

edited

Loading

mkavulich left a comment •

edited by mgduda

Loading