Effect of restart file and start date on HYCOM outputs

[Dianne Deauna]

Good day,

I have a question regarding the effect of the restart file and start dateon HYCOM model outputs. I am comparing the results from two separate runs which were conducted on the same server, using the same HYCOM version, relax and forcing files but had the following differences:

SEAb04.9

Start date: 2013_234 (August 22)

blkdat.input file:

number of days between model restart output = 9999.0

K-PROF output visc/diff coffs in archive = 0 (FALSE)

SEAb06.1

Start date: 2013_245 (September 2)

blkdat.input file:

number of days between model restart output = 15.25

K-PROF output visc/diff coffs in archive = 1 (TRUE)

When I compare the outputs for September 2, there are considerable differences between salinity, temperature, u and v-velocity. For bathymetry, there also seem to be small differences between the model runs. My question is, are these differences considered normal? If not, how could they be remedied? 

Thank you very much for your consideration.

[Dianne Deauna]

Good day,

I realize now that the restart file has nothing to do with it, since the model has not yet created a restart file for the time step I am comparing. Still, I wonder if these discrepancies between outputs are inherent in the model or if any steps can be taken to reduce them.

Thank you very much for your consideration.

--
--
You received this message because you are a member of HYCOM.org
To ask a question, send an email to fo...@hycom.org

---
You received this message because you are subscribed to the Google Groups "HYCOM.org Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to forum+un...@hycom.org.

--

Josephine Dianne L. Deauna

Research Assistant

Physical Oceanography Laboratory

Marine Science Institute, UP Diliman

(632) 9223957

[alan.wa...@hycom.org]

In general, ocean models are non-linear systems where a small difference in initial state can grow over time.   Usually it takes 2-3 months for the solutions to diverge significantly.

In your case, you did not say what "start date" means.  If you were starting from climatology - you still have climatology after 1 day but the model has a bit of time to adjust from Aug 22.  If you were starting from a data-assimilative state, this tells you that a 10 day forcast gives different results from a nowcast plus a 1 day forecast.

Restarts are designed so that starting from a restart gives *identical* results to the original run (e.g. with the restrart half way through).  However, since HYCOM restarts are real*4 this means that we deliberatly change resolution from real*8 to real*4 for all fields every time we write a restart.  So a run with 'rstrfq'=1.0 (say) will give slightly different results to one with 'rstrfq'=9999.0 (i.e. only at the end of each model run).

Alan.

[Dianne Deauna]

Good day Sir Alan,

Thank you very much for your reply. We do not start from climatology or data-assimilative states. The way we run the model is to create a 1-day restart file using negative limits, so for SEAb0.04, the restart day is August 21, while for SEAb06.1 the restart day is September 1. Should we consider running the model from climatology?

We did notice significant deviations around January 2014 (we consider the September to December 2013 runs as spin-up; please see attached). Given the real*8 to real*4 conditions, should we then consider these deviations as inherent in the model? We are particularly concerned with how temperature reacts, as the difference has a magnitude of +/- 4 degrees which can have a lot of implications (biological, physical, etc). 

Please note that initially we were trying to differentiate the effects of adding river discharge to the model, but given that even without river forcing there are significant differences between model runs, it becomes difficult to ascertain exactly how changing different forcing parameters can affect the model outputs.

 

[Dianne Deauna]

Hello,

I hope you don't mind, just to give a better picture of how SST varies and why we're concerned about it, I've attached a plot of the same data for temperature but with different color limits. Thank you!

[alan.wa...@hycom.org]

What do you mean by "1-day restart file using negative limits"?

Alan.

[Dianne Deauna]

Hello,

Basically we generate a 1-day model run using a limits file with a negative date.

For example, the contents of the file '061y113d244h00.limits' are:

-41152.00000    41153.00000

in order to generate a 1-day restart file which we can then use for the actual model run.

[Dianne Deauna]

Hello,

I would also like to provide additional information regarding the model runs, which might help in determining the cause for the SST discrepancies:

HYCOM version: 2.2.72 on 32 cores under ompi 

Resolution: 0.04 degrees

Bathymetry from ETOPO2v2

Relax from PHC climatology

Nesting from global 0.08 model

Real time forcing using 3 hourly NAVGEM 1.2_0.5a (wind, wind speed, SST, flux and precipitation)

Again, thank you very much for your consideration.

[alan.wa...@hycom.org]

A negative initial day in the limits file means "start from climatology".  For yrflag=3, it used to mean read the restart_in file but ignore a mismatch in model time but now it always means start from climatology.  Your .log file should make this clear.

This is almost certainly not what you want to do.

Typically for nested cases, we "cold start" from an outer model archive interpolated to the inner grid.  This generates gravity waves due the the mismatch in grids, so doing this a day early to clear out some of the noise makes sense (there will still be adjustment gravity waves after one day).

To generate a restart from an archive you need a "template" restart file. and it should be from the January climatology.  See https://groups.google.com/a/hycom.org/d/msg/forum/yqk5BRro0ro/Nq4KLiqiAwAJ

Alan.

[Dianne Deauna]

Thank you very much for your reply, we'll work on generating the climatology runs. May I ask what the latest relax and forcing files for climatology are recommended for use? And am I correct in assuming that a climatology run should not be run with nesting, and should have closed boundaries? 

[Dianne Deauna]

Yes, we do start from a "cold start" for our nested model. 

We did two simulation runs to compare the effect of the restart file. We assumed that "restarts are designed so that starting from a restart gives *identical* results to the original run", but this is not the case. The results from a run that creates a restart file every 15 days differ from a run that creates a restart file at the end of the simulation as shown in my previous attachments. 

[alan.wa...@hycom.org]

I don't understand why the 1st day in the limits file is negative if you are cold starting from an interpolated archive file.  As I said above, check that you are indeed reading in a restart file  by looking at the .log file.

In any case, the run with restarts every 15 days and the run with one restart at the end should have *identical* (cmp the .a files) archive files up until the 1st restart is written out.  They can then diverge, but I would not expect the divergence to be large initially.

Alan.

[Dianne Deauna]

The limits file where the 1st day is negative is used to make a 1-day run so we can get an interpolated archive file, and we then use the archive to restart routine to generate the restart file, which we then use to start the actual run. This is why the limits file, as I showed you is for September 1, but our actual run begins on September 2.

In our case, there is a divergence even without a restart file being generated. I'm reattaching the plots which show the difference in parameters. The first model run began on August 22, while the other model started on September 2. The plots shown are the difference between September 2 of the first and second run. Take note that the exact same parameters were used, the only difference being when restart files are generated. Again, I'd like to ask if this is inherent in running HYCOM.

With regards to the climatological run, I'd like to ask which latest relax and forcing files should be used. Also, how do we setup the model so that it runs for 10 years, but with the same conditions per month every year? 

Thank you very much for your consideration. 

[Till Rasmussen]

Hi Dianne

If you want to use the same restart file for all 10 years you need to change the time stamps in your .b files of your restart files. Then create nesting and forcing files for the new period.

Till