ocean import fields are zero when initialized from coldstart run #86
Comments
|
I don't think the cold2 step described above is true in my case as I
changed the run sequence to call write_restart before prep_ocean.
…On Wed, May 6, 2020 at 8:43 AM Denise Worthen ***@***.***> wrote:
The current implementation of S2S uses a 1 hour cold start followed by a
restart from the cold start. For ease of description, these will be
referred to as 'cold1' and 'cold2' below.
Using the NEMS mediator, the cold2 run is initializing the import state
for the ocean with all zero fields. This is can be verified by running
ufs-s2s-model with field dumping active. For the cold2 run, all fields in
the fields_ocn_import_(DATE)T00:00:00.nc are zero.
This behaviour is the result of the prep_ocn phase zero-ing the
accumulation fields and accumulation counters at the end of the phase. The
mediator restart files produced by the cold1 run
(mediator_FBaccumAtm_restart.*, mediator_FBaccumAtmOcn_restart.nc,
mediator_FBaccumIce_restart.nc) and the accumulation counts
(mediator_scalars_restart.txt) all contain zeros as a result.
When the cold2 run starts, the prep_ocn at the top of the run sequence
averages these accumulation fields (ie, divides by the accumulation count)
to create the import fields for the ocean. Since the fields are zero as
well as the counts, the import fields at time 00:00:00 are also zero.
This behaviour is replicated in the current CMEPS implmentation. However,
in CMEPS, the presence of a separate accum_avg phase (where the
accumulations are zero'd) and restart_write phase allows the run sequence
to be adjusted to produce valid (non-zero) mediator restarts for cold1.
This involves placing the med_phases_restart_write before of the
med_phases_prep_ocn_accum_avg in the cmeps cold1 run sequence.
For the NEMS mediator, this same change cannot be accomplished without
separating out the accumulation-related calls from prep_ocn.
A secondary issue w/rt the current cold1/cold2 sequence is that cold1 is
at the end of 2 ocean coupling steps. Even if the mediator restarts contain
non-zero values (ie, as can be accomplished with CMEPS), this would
seemingly result in import fields for the ocean at time 00:00:00 that are
in some sense forward by a coupling timestep relative to the ATM and ICE.
An alternative to the cold1/cold2 sequence is to use implement an ocean
lag startup. In this case, only the cold2 sequence is implemented but a
special setting is used so that the ocean model does not advance on the
first coupling timestep. Instead, the ocean model clock is set back to the
start time. On the second coupling timestep, the ocean model advances for 2
coupling timesteps. For example, if the coupling timestep is 30 minutes and
the internal ocean timesteps are 30 mins (thermo) and 15 mins (dynamic),
the model will step forward 2 thermo dynamic timesteps and 4 dynamic
timesteps. All subsequent coupling timesteps for the ocean model proceed as
normal.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#86>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ALLVRYWEDUQM26PTNND7BW3RQFLN7ANCNFSM4M2NDYYQ>
.
--
Dr. Shrinivas Moorthi
Research Meteorologist
Modeling and Data Assimilation Branch
Environmental Modeling Center / National Centers for Environmental
Prediction
5830 University Research Court - (W/NP23), College Park MD 20740 USA
Tel: (301)683-3718
e-mail: [email protected]
Phone: (301) 683-3718 Fax: (301) 683-3718
|
|
On the other hand, it is true for cold1; I guess it is defeating the reason
for cold1, which, as I understand, is to provide non zero atmospheric
forcing to the ocean through the mediator restart files.
On Wed, May 6, 2020 at 10:06 AM Shrinivas Moorthi - NOAA Federal <
[email protected]> wrote:
… I don't think the cold2 step described above is true in my case as I
changed the run sequence to call write_restart before prep_ocean.
On Wed, May 6, 2020 at 8:43 AM Denise Worthen ***@***.***>
wrote:
> The current implementation of S2S uses a 1 hour cold start followed by a
> restart from the cold start. For ease of description, these will be
> referred to as 'cold1' and 'cold2' below.
>
> Using the NEMS mediator, the cold2 run is initializing the import state
> for the ocean with all zero fields. This is can be verified by running
> ufs-s2s-model with field dumping active. For the cold2 run, all fields in
> the fields_ocn_import_(DATE)T00:00:00.nc are zero.
>
> This behaviour is the result of the prep_ocn phase zero-ing the
> accumulation fields and accumulation counters at the end of the phase. The
> mediator restart files produced by the cold1 run
> (mediator_FBaccumAtm_restart.*, mediator_FBaccumAtmOcn_restart.nc,
> mediator_FBaccumIce_restart.nc) and the accumulation counts
> (mediator_scalars_restart.txt) all contain zeros as a result.
>
> When the cold2 run starts, the prep_ocn at the top of the run sequence
> averages these accumulation fields (ie, divides by the accumulation count)
> to create the import fields for the ocean. Since the fields are zero as
> well as the counts, the import fields at time 00:00:00 are also zero.
>
> This behaviour is replicated in the current CMEPS implmentation. However,
> in CMEPS, the presence of a separate accum_avg phase (where the
> accumulations are zero'd) and restart_write phase allows the run sequence
> to be adjusted to produce valid (non-zero) mediator restarts for cold1.
> This involves placing the med_phases_restart_write before of the
> med_phases_prep_ocn_accum_avg in the cmeps cold1 run sequence.
>
> For the NEMS mediator, this same change cannot be accomplished without
> separating out the accumulation-related calls from prep_ocn.
>
> A secondary issue w/rt the current cold1/cold2 sequence is that cold1 is
> at the end of 2 ocean coupling steps. Even if the mediator restarts contain
> non-zero values (ie, as can be accomplished with CMEPS), this would
> seemingly result in import fields for the ocean at time 00:00:00 that are
> in some sense forward by a coupling timestep relative to the ATM and ICE.
>
> An alternative to the cold1/cold2 sequence is to use implement an ocean
> lag startup. In this case, only the cold2 sequence is implemented but a
> special setting is used so that the ocean model does not advance on the
> first coupling timestep. Instead, the ocean model clock is set back to the
> start time. On the second coupling timestep, the ocean model advances for 2
> coupling timesteps. For example, if the coupling timestep is 30 minutes and
> the internal ocean timesteps are 30 mins (thermo) and 15 mins (dynamic),
> the model will step forward 2 thermo dynamic timesteps and 4 dynamic
> timesteps. All subsequent coupling timesteps for the ocean model proceed as
> normal.
>
> —
> You are receiving this because you are subscribed to this thread.
> Reply to this email directly, view it on GitHub
> <#86>, or
> unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ALLVRYWEDUQM26PTNND7BW3RQFLN7ANCNFSM4M2NDYYQ>
> .
>
--
Dr. Shrinivas Moorthi
Research Meteorologist
Modeling and Data Assimilation Branch
Environmental Modeling Center / National Centers for Environmental
Prediction
5830 University Research Court - (W/NP23), College Park MD 20740 USA
Tel: (301)683-3718
e-mail: ***@***.***
Phone: (301) 683-3718 Fax: (301) 683-3718
--
Dr. Shrinivas Moorthi
Research Meteorologist
Modeling and Data Assimilation Branch
Environmental Modeling Center / National Centers for Environmental
Prediction
5830 University Research Court - (W/NP23), College Park MD 20740 USA
Tel: (301)683-3718
e-mail: [email protected]
Phone: (301) 683-3718 Fax: (301) 683-3718
|
|
Update this issue to include next steps:
|
|
I added a state_diagnose function to the mom6 nuopc cap and turned on the FB_diagnose in mep_phases_prep_ocn_avg. Using these two features, I have verified that:
Three test cases corresponding to these items were made. At hour 6, the difference in SST from the initial MOM6 SST for the three cases is shown:
|
|
Thanks for testing and making these plots, Denise. Can you make a plot of
fig3(ocean lag)-sst in current 2day run with mediator fields from cold
start? Also will the ocean always lag one slow step when we remove the cold
start?
…On Mon, Aug 3, 2020 at 8:15 AM Denise Worthen ***@***.***> wrote:
I added a state_diagnose function to the mom6 nuopc cap and turned on the
FB_diagnose in mep_phases_prep_ocn_avg. Using these two features, I have
verified that:
1.
Using the existing cold start procedure produces ocean import fields
of 0.0 for the first ocean coupling timestep when restarting from coldstart.
2.
Moving the restart write to earlier in the cold run sequence (as
suggested by @mvertens <https://github.com/mvertens>) produces
non-zero import fields for the ocean at the first coupling timestep.
3.
Using the FV3 fix provided by @junwang-noaa
<https://github.com/junwang-noaa> to provide the bottom-most atm
values to the ice at the first coupling timestep and using an ocean-lag for
the first ocean coupling timestep allows the model to run in a concurrent
run sequence from ICs with no cold start mediator file required.
Three test cases corresponding to these items were made. At hour 6, the
difference in SST from the initial MOM6 SST for the three cases is shown:
[image: sst_diffs]
<https://user-images.githubusercontent.com/40498404/89181250-50102200-d561-11ea-8725-3b4f5406793e.jpg>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#86 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TJYLHBCWSSDRSAT2P3R62S4NANCNFSM4M2NDYYQ>
.
|
|
Yes, my test cases are on Cheyenne and I ran 48 hours. Do you want the ocean lag SST (case3)- cold start SST (case1) at 48 hours? The ocean lag is only on the first coupling timestep. On the first coupling timestep, the ocean does not advance. On the second coupling timestep, MOM6 advances by two timesteps. After that, MOM6 advances normally. The relevant code can be seen LN 1382 |
|
I'd like to see the sst difference between no cold start and with cold
start run.
From the code, MOM6 will run with 2*ocean_timestep first then go back to
normal ocean timestep. Is "runtype" set to true for the first two run time
steps? I seem not able to find that in the code
…On Mon, Aug 3, 2020 at 9:02 AM Denise Worthen ***@***.***> wrote:
Yes, my test cases are on Cheyenne and I ran 48 hours. Do you want the
ocean lag SST (case3)- cold start SST (case1) at 48 hours?
The ocean lag is only on the first coupling timestep. On the first
coupling timestep, the ocean does not advance. On the second coupling
timestep, MOM6 advances by two timesteps. After that, MOM6 advances
normally.
The relevant code can be seen LN 1382
<https://github.com/NOAA-EMC/MOM6/blob/728c429c32b49b192132e00ccd0e12c45ff0b43d/config_src/nuopc_driver/mom_cap.F90#L1382>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#86 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TIQWDEFC4KDTEKE3ITR62YNFANCNFSM4M2NDYYQ>
.
|
|
Runtype is set in the mom_cap at InitializeAdvertise. It is read from nems.configure from the variable "start_type". For non-cesm, it will be set as "initial". We are currently not really making use of the start_type variable since our set up uses the namelists of components to set whether we're doing an initial vs continue run. We do need to be careful during actual model restarts ("warm start") that we make consistent use of runtype. |
|
I see, thanks
…On Mon, Aug 3, 2020 at 9:44 AM Denise Worthen ***@***.***> wrote:
Runtype is set in the mom_cap at InitializeAdvertise. It is read from
nems.configure from the variable "start_type". For non-cesm, it will be set
as "initial".
We are currently not really making use of the start_type variable since
our set up uses the namelists of components to set whether we're doing an
initial vs continue run. We do need to be careful during actual model
restarts ("warm start") that we make consistent use of runtype.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#86 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TLJDEZR4D6ADMNOS63R625J7ANCNFSM4M2NDYYQ>
.
|
|
This is the SST difference at 48 hours for the 20120101 ICS between using the ocean lag and using the original cold start mediator file.
|
This was referenced Aug 22, 2020
DeniseWorthen
added a commit
that referenced
this issue
Sep 8, 2020
* Replaces CICE5 with CICE6: Issue #159 * Adds required CICE6 items to nems.configure. Removes non-cmeps nems.configure files. Updates default_vars and edit_inputs to allow CICE6 resolution and processor count to be specified at run time. Updates ice_in_template for CICE6 * Updates to esmf8.1.0bs27 to allow ICE and WAV on separate PEs (Issue #172) * Adds cpl_mode (Issue #138) and use_coldstart as new configurable variables (Issues #86, #53). * Fixes coldstart run sequence by calling the restart write phase before the accum avg phase (Issue #167)
|
Closing; this issue is resolved by the commit of the 1step coldstart PR #187 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The current implementation of S2S uses a 1 hour cold start followed by a restart from the cold start. For ease of description, these will be referred to as 'cold1' and 'cold2' below.
Using the NEMS mediator, the cold2 run is initializing the import state for the ocean with all zero fields. This is can be verified by running ufs-s2s-model with field dumping active. For the cold2 run, all fields in the fields_ocn_import_(DATE)T00:00:00.nc are zero.
This behaviour is the result of the prep_ocn phase zero-ing the accumulation fields and accumulation counters at the end of the phase. The mediator restart files produced by the cold1 run (mediator_FBaccumAtm_restart.*, mediator_FBaccumAtmOcn_restart.nc, mediator_FBaccumIce_restart.nc) and the accumulation counts (mediator_scalars_restart.txt) all contain zeros as a result.
When the cold2 run starts, the prep_ocn at the top of the run sequence averages these accumulation fields (ie, divides by the accumulation count) to create the import fields for the ocean. Since the fields are zero as well as the counts, the import fields at time 00:00:00 are also zero.
This behaviour is replicated in the current CMEPS implmentation. However, in CMEPS, the presence of a separate accum_avg phase (where the accumulations are zero'd) and restart_write phase allows the run sequence to be adjusted to produce valid (non-zero) mediator restarts for cold1. This involves placing the med_phases_restart_write before of the med_phases_prep_ocn_accum_avg in the cmeps cold1 run sequence.
For the NEMS mediator, this same change cannot be accomplished without separating out the accumulation-related calls from prep_ocn.
A secondary issue w/rt the current cold1/cold2 sequence is that cold1 is at the end of 2 ocean coupling steps. Even if the mediator restarts contain non-zero values (ie, as can be accomplished with CMEPS), this would seemingly result in import fields for the ocean at time 00:00:00 that are in some sense forward by a coupling timestep relative to the ATM and ICE.
An alternative to the cold1/cold2 sequence may be to implement an ocean lag startup. In this case, only the cold2 sequence is implemented but a special setting is used so that the ocean model does not advance on the first coupling timestep. Instead, the ocean model clock is set back to the start time. On the second coupling timestep, the ocean model advances for 2 coupling timesteps. For example, if the coupling timestep is 30 minutes and the internal ocean timesteps are 30 mins (thermo) and 15 mins (dynamic), the model will step forward 2 thermo dynamic timesteps and 4 dynamic timesteps. All subsequent coupling timesteps for the ocean model proceed as normal.
The text was updated successfully, but these errors were encountered: