Branch runs fail at double allocation of FATES structures

[aldivi]

I have a local "fix" for this in my harvest branch aldivi/elm/fates-harvest in the aldivi/fates github fork. I am not sure how this relates to the rest of the host model restart code, or to the best way to start runs from other runs.

In clm_initiallizeMod (line 914) alm_fates%init_coldstart is called if is_restart() is false. But is_restart() is false for branch runs, which act as restarts. So the both coldstart and restart functions allocate FATES structures.
Locally, I added an additional .and. condition for coldstart: nsrest /= nsrBranch

Likewise, in clmfates_interfaceMod (line 310), pass_is_restart passes the is_restart() value to FATES, which is then used by set_site_properties to determine whether to set some initial values or not (if not, later the values are set from a restart file).
Locally, I added the condition: .or. nsrest .eq. nsrBranch to pass a restart flag if it is a branch run.

[ckoven]

@aldivi does this happen if you do a hybrid run too? I just realized (because @jkshuman told me) that a hybrid run is necessary if turning on fire using a restart from a case where fire is off, and I managed to get it to work; normally I've always just used the finidat field in the user_nl_clm file. so does a hybrid run work for you?

[aldivi]

@ckoven I haven't tried a hybrid run, but I don't think this happens for hybrid runs because they are flagged as startup runs, so they shouldn't trigger the reading/initializing of restart files/structures. One usually (always?) provides an initial file (via finidat through run_refcase and run_refdate) for a hybrid run so that it isn't a cold start. In the code there generally isn't a distinction between a hybrid run and a startup run, just a distinction as to whether or not there is a finidat file. I think there are some namelist checks for things that can or cannot be set with hybrid vs startup, but some of these are just superficial restrictions such as a cold start cannot be set if finidat is set (not sure why two separate run types are needed for this). I think that runs starting from the end of spinups are always configured as a hybrid run with an initial conditions file.

[glemieux]

@aldivi is this still a live issue?

[aldivi]

@glemieux
these issues are actually in the host model code. I think i put them here as they might have pertained to both clm and elm.
this is not live anymore for elm. the fixes i noted above are in the elm master.
i don't know about clm.

[glemieux]

Thanks for the clarification. CTSM still uses the same checks as far as I can tell, so this should be addressed. For others, here is the e3sm PR that includes the updates @aldivi applied: E3SM-Project/E3SM#4024

• Check branch run failure for clm-fates