selection of float layer failed, then stop integration

[Daniel Moreira]

Hi,

I've got an error involving floats that stop the integration, since floats storage has failed and so mean SSH and other means goes to NaN. The storage error goes through many floats (257~4900).
This isn't the first year of integration, so it is not supposed to be a float configuration problem.
See below if someone can help me.


storing float parameters
 14129025 (2008/166 07) mean      SSH (mm):  399.26  ( 7.6E+01 to  6.5E+02)
 14129025 (2008/166 07) region-wide mean KE:     18324.3384807523
 14129025 (2008/166 07) region-wide mean APE:      842.1446975281
 step 14129025 day 39247.29 -- archiving completed --
 storing float parameters
 warning - selection of float layer failed
float   257 aground, time step 14129040 float time step    59880
   ier   0 exit point  4 ngood 1-4 16 16 15
...
...
...
14129040 (2008/166 08) mean      SSH (mm):     NaN  (-1.1E+02 to  6.7E+02)
  
 error - NaN or Inf detected  
  
 14129040 (2008/166 08) region-wide mean KE:                  NaN
 14129040 (2008/166 08) region-wide mean APE:                 NaN

**************************************************
 (hycom)
 **************************************************

 mpi_finalize called on processor         


Thanks in advance,
Daniel

[Alan]

HYCOM checks for problems (negative layer thickness) every 3 time steps, and checks for NaN only when writing to the .log file (in your case every model hour).  So it is possible that the model has a "normal" blow up and the float routine detected the problem before the standard model checks caught it.  The easiest way to tell would be to rerun the last segment without floats.  If it is the model blowing up, the usual fix is to halve the baroclinic time step for one model seqment.  Leave the barotropic time step as large as possible.  You could also just try the smaller time step with floats still on.

Alan.

[Daniel Moreira]

Alan,

Thank you for your first analysis. I have been analysing some data and I would like to share some results:

1- It wasn't a float issue. You were correct.
2- It's not a chronic problem, since SSH fields and dp and u_vel merional section were corrected until day 166_00. The problem occur at hour 08 of day 166.

Take a look on these SSH field of hour 07 and 08 and dp meridional section of hour 08 in 166th day.

I have already checked the forcing fields as well.

ps.: I don't think it was a problem wit baroclinic and barotropic time step since the experiment have been run for years.

Thanks in advance,
Daniel

[Alan]

Is the field NaN in the missing box?  I am concerned that it is a rectangle, I would expect an expanding disturbance to be nearly circular.

My 1st suspect is the forcing, but you say it is ok (did you use hycom_NaN to check there are no NaN's in the forcing?  Note that min,max stats typically don't catch these).

Always try reducing the 'baclin' time step (e.g. to half its usual value) when you hit a problem, this often gets the run past the problem.  Keep the 'batrop' time step as large as possible.

Alan.