Predicted values above maximum [Error]

[isaacaka]

This is an error which occurs when the value predicted by the ML method is far beyond the range in value that NEMO accepts.

E.g in this case the predicted SSH is 500m whereas the maximum that SSH can be is 20m

  ===>>> : E R R O R
 
          ===========
 
   stp_ctl: |ssh| > 20 m  or  |U| > 10 m/s  or  S <= 0  or  S >= 100  or  NaN en
 counter in the tests
 
 kt 1152001 |ssh| max   518.5     at i j   37 155    MPI rank 7
 kt 1152001 |U|   max   46.04     at i j k 60 183  1 MPI rank 7
 kt 1152001 Sal   min   34.59     at i j k  3 152 10 MPI rank 6
 kt 1152001 Sal   max   36.78     at i j k  3  86 11 MPI rank 2
 
        ===> output of last computed fields in output.abort* files
 
                     iom_close ~~~ close file: ./DINO_pr2_01152000_restart.nc ok

AAAAAAAA

 
  ===>>> : E R R O R
 
          ===========
 
 
     ==>>>   nemo_gcm: a total of            1  errors have been found
 
             Look for "E R R O R" messages in all existing ocean_output* files

[ma595]

I also ran into the same issue when attempting to restart NEMO (trained on 50 years of data):

 iom_close ~~~ close file: output.abort_0000.nc ok

  ===>>> : E R R O R

          ===========

   stp_ctl: |ssh| > 20 m  or  |U| > 10 m/s  or  S <= 0  or  S >= 100  or  NaN en
 counter in the tests

 kt 576001 |ssh| max   45.81     at i j   43 169    MPI rank 32
 kt 576001 |U|   max   10.43     at i j k  2 147  9 MPI rank 24
 kt 576001 Sal   min   34.68     at i j k 15  30 10 MPI rank  3
 kt 576001 Sal   max   37.88     at i j k  2  58  2 MPI rank  9

        ===> output of last computed fields in output.abort* files

                     iom_close ~~~ close file: ./NEW_DINO_00576000_restart.nc ok

AAAAAAAA


  ===>>> : E R R O R

          ===========


     ==>>>   nemo_gcm: a total of            1  errors have been found

             Look for "E R R O R" messages in all existing ocean_output* files

[isaacaka]

Looks like this issue is at least in part related to the code forecasting an incorrect and longer number of years than we intend.

e.g training on 30 years and forecasting 60 years (incorrectly) instead of the 30 years we intend.
See issue: #43

Dimensionality reduction and forecasting techniques are also other factors.