HAM cleanup for flexible operator splitting strategy
Posted by Sylvaine Ferrachat 02/10/2022 https://redmine.hammoz.ethz.ch/issues/818
ECHAM and ICON with ECHAM-physics using lparamcpl=.true. apply “operator strang splitting”, i.e. every process uses as the state an updated state from all previous processes. The reason is to avoid negative tracer concentrations (a process can’t remove what a previous process had already removed). The disadvantage is that the order of processes is important.
With lparamcpl=.false.
in ICON with ECHAM-physics all tendencies are collected and added only in the end. I.e. all processes use the same state. This is only compatible with small timesteps.
In ECHAM the “operator strang splitting” was implemented such that the fields were not updated during the model time steps but each process contributed to the tendencies and therefore a current “intermediate” state was used for each process by adding the current tendencies to the fields. In ICON this was changed (ToDo: date / version to specify here): with lparamcpl=.true.
the fields AND the tendencies are updated after each process and each process takes the fields as the current state (NOT field + tendencies).
All physics routines were accordingly adapted.
--> HAM needs such an adaptation as well. This is the purpose of this issue
Important note: icon-hammoz-distrib-prerelease_1.0 is not affected by this problem, as the lparamcpl feature was introduced in a newer icon version than the one on which icon-hammoz-distrib prerelease_1 is based.