clm bug 2082 - problem with ED history files on restarts

[bandre-ucar]

Summary of Issue:

Original bug: CLM - 2082

Need to test if this is still relevant.

Mariana Vertenstein 2014-11-02 14:26:06 MST
I have found the that comparing the first ED history files upon restarts - they are not bit-for-bit - even though the model remains bit-for-bit and the restart test passes. In particular - for clm4_5_1_r091. As an example - if I look at the case 

ERS_D_Mmpi-serial.1x1_brazil.ICLM45CNED.yellowstone_gnu.clm-edTest

and write 1d history files daily  - I will see that upon restart (which starts from the beginning of day 7)

clm2.h0.0001-01-08-00000.nc is different between the initial and restart runs in the following fields:

 RMS GPP                              2.2001E-04            NORMALIZED  3.1505E-02
 RMS LITTER_OUT                       4.2088E+00            NORMALIZED  2.8284E+00
 RMS NPP                              1.1212E-04            NORMALIZED  3.2232E-02

However clm2.h0.0001-01-12-00000.nc are identical between the initial and restart runs.

The same thing holds for clm4_5_1_r091 with the koven refactoring  changes I have brought in. However, in this case there are now only 2 fields that are different.

 RMS GPP                              2.2001E-04            NORMALIZED  3.1505E-02
 RMS NPP                              1.1212E-04            NORMALIZED  3.2232E-02

The LITTER_OUT problem is not longer there - but the other two are still persistent.

This should be fixed - but it also points to the need to compare history files upon restart as well.

[rgknox]

So if I understand correctly, it looks as though the state variables that dictate the GPP and NPP diagnostics are restarting correctly, but for some reason the diagnostics on that first file are not correct.

So apparently it has something to do with how those history variables are initialized?

[bandre-ucar]

That is my understanding. I was wondering if it had something to do with #24 or one of the other history bug fixes we've made. Best thing to do is test if the problem still exists.

[ckoven]

this was solved a while ago; closing.